TYPE Review PUBLISHED 15 November 2022 DOI 10.3389/fmicb.2022.1006659 Advances and challenges in OPEN ACCESS automated malaria diagnosis using digital microscopy imaging EDITED BY Jun Feng, Shanghai Municipal Center for Disease Control and Prevention (SCDC), China REVIEWED BY with artificial intelligence tools: A review Adugna Woyessa Gemeda, Ethiopian Public Health Institute, Ethiopia Denise Anete de Alvarenga, Fiocruz Minas, IRR, Brazil *CORRESPONDENCE Carles Rubio Maturana 1,2, Allisson Dantas de Oliveira 3, Sergi Joan Joseph-Munné joan.joseph@vallhebron.cat Nadal 4, Besim Bilalli 4, Francesc Zarzuela Serrat 1, Mateu Espasa Elisa Sayrol Clols Soley 2,5, Elena Sulleiro Igual 1,2,6, Mercedes Bosch 7, Anna Veiga elisa.sayrol@upc.edu Lluch 7, Alberto Abelló 4, Daniel López-Codina 3, Tomàs SPECIALTY SECTION This article was submitted to Pumarola Suñé 1,2, Elisa Sayrol Clols 8* and Joan Infectious Agents and Disease, a section of the journal Frontiers in Microbiology Joseph-Munné 1* RECEIVED 29 July 2022 1 Microbiology Department, Vall d’Hebron Research Institute, Vall d’Hebron Hospital Campus, ACCEPTED 26 September 2022 Barcelona, Spain, 2 Universitat Autònoma de Barcelona (UAB), Barcelona, Spain, 3 Computational PUBLISHED 15 November 2022 Biology and Complex Systems Group, Physics Department, Universitat Politècnica de Catalunya (UPC), Castelldefels, Spain, 4 Data Base Technologies and Information Group, Engineering Services CITATION and Information Systems Department, Universitat Politècnica de Catalunya (UPC), Barcelona, Spain, Rubio CM, de Oliveira AD, Nadal S, Bilalli B, 5 Clinical Laboratories, Microbiology Department, Hospital Universitari Parc Taulí, Sabadell, Spain, Zarzuela FS, Espasa MS, Sulleiro E, Bosh M, 6 CIBERINFEC, ISCIII- CIBER de Enfermedades Infecciosas, Instituto de Salud Carlos III, Madrid, Veiga ALl, Abelló A, López-Codina D, Spain, 7 Probitas Foundation, Barcelona, Spain, 8 Image Processing Group, Telecommunications and Pumarola TS, Sayrol EC and Signal Theory Group, Universitat Politècnica de Catalunya (UPC), Barcelona, Spain Joseph-Munne J (2022) Advances and challenges in automated malaria diagnosis using digital microscopy imaging with artificial intelligence tools: A review. Malaria is an infectious disease caused by parasites of the genus Plasmodium Front. Microbiol. 13:1006659. spp. It is transmitted to humans by the bite of an infected female Anopheles doi: 10.3389/fmicb.2022.1006659 mosquito. It is the most common disease in resource-poor settings, COPYRIGHT with 241 million malaria cases reported in 2020 according to the World © 2022 Rubio, de Oliveira, Nadal, Bilalli, Zarzuela, Espasa, Sulleiro, Bosh, Veiga, Health Organization. Optical microscopy examination of blood smears Abelló, López-Codina, Pumarola, Sayrol is the gold standard technique for malaria diagnosis; however, it is a time- and Joseph-Munne. This is an open-access consuming method and a well-trained microscopist is needed to perform article distributed under the terms of the Creative Commons Attribution License (CC the microbiological diagnosis. New techniques based on digital imaging BY). The use, distribution or reproduction in analysis by deep learning and artificial intelligence methods are a challenging other forums is permitted, provided the alternative tool for the diagnosis of infectious diseases. In particular, systems original author(s) and the copyright owner(s) are credited and that the original based on Convolutional Neural Networks for image detection of the malaria publication in this journal is cited, in parasites emulate the microscopy visualization of an expert. Microscope accordance with accepted academic practice. No use, distribution or automation provides a fast and low-cost diagnosis, requiring less supervision. reproduction is permitted which does not Smartphones are a suitable option for microscopic diagnosis, allowing image comply with these terms. capture and software identification of parasites. In addition, image analysis techniques could be a fast and optimal solution for the diagnosis of malaria, tuberculosis, or Neglected Tropical Diseases in endemic areas with low resources. The implementation of automated diagnosis by using smartphone applications and new digital imaging technologies in low-income areas is a challenge to achieve. Moreover, automating the movement of the microscope slide and image autofocusing of the samples by hardware implementation would systemize the procedure. These new diagnostic tools would join the Frontiers in Microbiology 01 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 global effort to fight against pandemic malaria and other infectious and poverty-related diseases. KEYWORDS malaria diagnosis, digital imaging techniques, deep learning, artificial intelligence, microscopic examination, smartphone application, malaria Introduction The implementation of early detection systems for malaria epidemics is a high priority in Sub-Saharan African regions Malaria is one of the most common infectious diseases (Guintran et al., 2006). New advances in the regulation and worldwide. It is caused by Plasmodium parasites and transmitted development of malaria vaccines, such as the RTS, S/ASO1 to humans by the bite of an infected female mosquito of the vaccine recommendation by the WHO, can reinvigorate the fight Anopheles genus. Over 241 million malaria cases were estimated against malaria (WHO recommends groundbreaking malaria in 2020, an increase from the 227 million of 2019 according to the vaccine for children at risk, 2021). Laboratory techniques for World Health Organization (WHO) (World Malaria Report WHO malaria diagnosis by detecting Plasmodium parasites are 2021). Malaria is endemic in 85 countries and caused 627,000 extensively used worldwide; microscopic visualization of thin and deaths in 2020. Africa is the most affected continent with 95% of thick blood smears is the gold standard technique for malaria all malaria cases reported and 96% of all deaths (Talapko et al., diagnosis. RDTs are also used as recommended diagnostic tools 2019; World Malaria Report WHO 2021). Low-income countries and could be an affordable complement for a precise diagnosis due with non-accessible healthcare resources are the most affected to their rapidness and easy handling. Both microscopic regions and malaria-related mortality has a high correlation with visualization and RDTs have their limitations and new diagnostic poverty rates (Ren, 2019). Socioeconomic data were collected in techniques are emerging to complement the tools used nowadays. several studies to demonstrate the aforementioned correlation, As a breakthrough, new image analysis techniques based on deep describing the global health situation of malaria in low-income learning, a subfield in artificial intelligence (AI), are being countries (Ricci, 2012; Konishi et al., 2016). An early diagnosis, developed for the automated diagnosis of blood slides. suitable treatment, and prevention strategies such as vaccination Distinguishing between erythrocytes infected or uninfected with or mosquito net control are crucial to fighting the infection. Due malaria parasites is possible with deep learning detection-based to its high global health impact, this infectious disease is still a models. Image analysis techniques allow the detection of malaria global issue. In addition, the COVID-19 pandemic has increased parasites in digital images by pre-trained deep learning models the number of malaria deaths and cases from previous years, due with large image datasets. This process would emulate the optical to the high impact of this pandemic on the administration of microscope visualization of thick and thin blood smear samples healthcare resources worldwide (Heuschen et al., 2021). and automate the procedure. Smartphone applications could Plasmodium infection is produced by several protozoan integrate image analysis technology based on AI and would be an parasites of the genus Plasmodium spp. (Tangpukdee et al., 2009). affordable option for resource-poor environments in Five species of malaria cause infection in humans: P. falciparum, endemic areas. P. vivax, P. ovale, P. malariae, and P. knowlesi. P. falciparum is the Identification of the different parasite morphologies in the most virulent species and produces the vast majority of deaths whole Plasmodium life cycle is crucial to perform a correct from severe malaria (Heide et al., 2019). The life cycle of diagnosis by microscopic examination of blood smears. The life Plasmodium parasites is represented in Figure 1. The life and cycle must be considered when experts perform manual labelling infective cycle of the five species are similar, and their morphology of digital images. Immature P. falciparum trophozoites (ring and biology are analogous (Talapko et al., 2019). stage), White Blood Cells (WBCs), and erythrocytes are Malaria treatment is crucial to reducing mortality. Prompt commonly labelled in malaria thick and thin blood smear digital treatment is recommended, within 24 h of the onset of fever, and images (Manescu et al., 2020). The labelled data would be used to is fundamental for the reduction of mortality among children train deep neural network models and create AI algorithms <5 years of age (Simba et al., 2018). After confirmation of capable of detecting parasites and cells. Plasmodium infection by laboratory diagnostic techniques, such as Rapid Diagnostic Tests (RDT) or microscopy, anti-malarial drugs are administered. The treatment used should be determined Malaria diagnosis by Plasmodium species, parasitaemia density, drug-resistant pattern where the infection was acquired, signs of severe malaria, Malaria diagnosis is crucial to treat and eradicate Plasmodium and patient tolerance of oral medication (Griffith et al., 2007). infections. An early diagnosis is determinant in effectively fighting Frontiers in Microbiology 02 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 FIGURE 1 Life cycle of the Plasmodium parasite. Mosquito, human liver, and human blood stages are represented. (1) Mosquito injects sporozoites; (2) Infected hepatocyte; (3) Ruptured exo-erytrocytic schizont; (4) Merozoites in peripheral blood; (5) Immature trophozoites in peripheral blood; (6) Ruptured schizont; (7) Some immature trophozoites develop into sexual precursor cells named gametocytes; (9) Ookinete (motile zygote). against infection. Laboratory diagnosis is accepted worldwide and Clinical symptomatology should be complemented with recommended for malaria detection (Tangpukdee et al. 2009; laboratory diagnostic techniques to confirm the presence of World Malaria Report WHO, 2021). Diagnostic methods for Plasmodium parasites. Blood smear samples are used in the vast infectious diseases should be fast, accurate, simple, and affordable majority of diagnostic techniques. (Vila et al., 2017). Several techniques are available and used to directly or indirectly detect the presence of malaria parasites in blood. Table 1 shows the advantages and disadvantages of the Microscopic examination of blood most important diagnostic methods for malaria parasite detection. smears Direct microscopic examination of blood smears to observe Clinical diagnosis malaria parasites is the gold standard technique for malaria diagnosis (Guintran et al., 2006; Collins and Jeffery, 2007; Heide Clinical diagnosis is the least expensive option for malaria et al., 2019). Prior to examination, the specimen is mostly stained diagnosis (Wongsrichanalai et al., 2007), although the non-specific with Giemsa or Leishman staining (Bejon et al., 2014), to afford symptomatology and possible confusion with other infections or the parasites a distinctive appearance (Malaria diagnosis and diseases with similar manifestations could overlap with the final treatment CDC, 2019). The protocol for Giemsa staining of diagnosis. Patient origin, malaria season, and age group are malaria blood films is a simple and fast technique to visualize the important aspects to consider. Clinical symptomatology could active form of parasites in blood (Turrientes and López, 2016). vary depending on the phase of the disease and the Plasmodium Malaria microscopy standard operating procedure is the protocol parasite species producing the infection. Plasmodium infection recommended by the WHO (Giemsa staining of malaria blood could produce asymptomatic, placental, uncomplicated, and films WHO, 2016). The blood smear examination procedure is severe malaria depending on the symptomatology and infection shown in Figure 2. phase (Molyneux, 1989; Bartoloni and Zammarchi, 2012; Phillips Knowing the life cycle of Plasmodium parasites (Figure 1) is et al., 2017). important to perform a correct identification of the different Frontiers in Microbiology 03 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 TABLE 1 Advantages and disadvantages of malaria diagnostic techniques. Diagnostic technique Advantages Disadvantages References Microscopic examination (i) Availability (i) Requires expert personnel Dowling and Shute (1966), Collins and Jeffery, (ii) Low-cost diagnosis (ii) Results are expert-dependent (2007), Guintran et al. (2006), Wangai et al. (2011), (iii) Parasite level calculations Poostchi et al. (2018), Heide et al. (2019), Malaria (iv) Species identification diagnosis and treatment CDC, 2019 Quantitative Buffy Coat (QBC) (i) Fast preparation and diagnosis (i) Requires expert personnel QBC Malaria Test, 2007, Tangpukdee et al. (2009), results (ii) Requires fluorescent microscopy Siciliano and Alano (2015), About Malaria (ii) High sensitivity (iii) Specialized instrumentation CDC, 2019, Shujatullah et al., 2006 Rapid Diagnostic Tests (RDTs) (i) Fast preparation and diagnosis (i) pfHRP2/3 gene deletions Wongsrichanalai et al. (2007), Gillet et al. (2009), results (ii) Low sensitivity with low parasite levels Murray and Bennett (2009), Tangpukdee et al. (ii) Easy handling (iii) Low sensitivity with P. ovale and P. (2009), Bejon et al. (2014), Nima et al. (2017), (iii) Low-cost diagnosis malariae species. Orish et al. (2018), Cunningham et al. (2019), (iv) Species identification (usually (iv) Cross-reactivity Response plan to phrp2 gene deletions (WHO, 2019), P. falciparum from non-P. (v) Prozone effect Ajakaye and Ibukunoluwa (2020), Kavanaugh et al. falciparum species) (2021), Kavanaugh et al. (2021) PCR (i) High sensitivity and specificity (i) Specialized instrumentation Johnston et al. (2006), Li et al. (2014), Poostchi (ii) Species identification (ii) Difficult implementation in endemic areas et al. (2018), Siwal et al. (2018), Haanshuus et al. (iii) Reference tool for comparative (iii) Expensive diagnosis (2019), Eshag et al. (2020), Leski et al. (2020), studies Feufack-Donfack et al. (2021), Mwenda et al. (2021) LAMP (i) High sensitivity and specificity (i) Specialized instrumentation Ocker et al. (2016), Selvarajah et al. (2020), Morris (ii) Species identification (ii) Expensive diagnosis and Aydin-Schmidt (2021) (iii) No thermocyclers needed Serology (i) Seroprevalence (i) Non-reliable diagnostic technique Tangpukdee et al. (2009), Oviedo et al. (2020) (ii) Malaria transmission (ii) Not indicative of active infection Flow cytometry (i) Quantification of infected (i) Low sensitivity Poostchi et al. (2018), Khartabil et al. (2022) erythrocytes (ii) Specialized instrumentation (ii) Automated parasite level (iii) Difficult implementation in endemic areas calculations Biomarkers (i) High diagnostic potential (i) Specialized instrumentation Jain et al. (2014), Krampa et al. (2017)) (ii) Easy handling QBC: Quantitative Buffy Coat, RDTs: Rapid Diagnostic Tests, PCR: Polymerase Chain Reaction, LAMP: Loop-Mediated Isothermal Amplification, Serology, Flow cytometry, and biomarkers. developmental stages of the parasites and the species for diagnosis. development asexual cycle, P. knowlesi infection can rapidly P. falciparum usually causes higher parasite levels and produces progress into severe malaria. Ring stage forms of P. knowlesi most malaria deaths in Africa (World Malaria Report WHO, resemble P. falciparum and mature trophozoites and schizonts are 2021). Maurer dots, poly-infected erythrocytes, and the similar to P. malariae forms (Amir et al., 2018). Gametocytes, the characteristic banana shape of gametocytes are distinctive traits of sexual stage of the parasite, are not responsible for clinical P. falciparum infection (Zekar and Sharman, 2021). P. vivax and symptoms (Treatment of malaria CDC, 2013). P. ovale are species sharing some similarities in the shape of Microscopic visualization of thin blood smears allows the parasites and quiescent liver forms. Both species infect young Plasmodium species identification from erythrocyte morphology erythrocytes, have Schüffner’s dots, tend not to have multiple rings and the distinctive features depending on the type of specimen per cell, and contain malarial pigment. P. malariae usually causes infection. Thick blood smears are more efficient and provide lower parasite levels, due to its 72-h development cycle (24 h higher sensitivity than thin blood smears (Wangai et al., 2011). The longer than P. falciparum and P. vivax), the lower production of combination of both methods allows experts to determine the type merozoites per erythrocytic cycle, the predilection of parasites to and severity of the infection with a precise diagnosis. Parasite level develop inside old erythrocytes and the earlier development of calculations are performed manually in both types of samples. immunity due to the combination of the previous factors (Collins Direct microscopy observation is a tedious and time-consuming and Jeffery, 2007). P. knowlesi is mostly present in Southeast Asia technique that requires experience and training. Continuous and was originally known as simian malaria. Due to its 24-h visualization of blood smears could trigger diagnostic errors due Frontiers in Microbiology 04 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 FIGURE 2 Blood smear microscopic examination procedure. Thick blood smear is first examined to determine the presence of malaria parasites. If the sample is positive, a thin blood smear is examined to determine Plasmodium species identification. Parasite quantification is performed to determine severity of the infection. to the difficulty of the procedure (Dowling and Shute, 1966). The concentration of parasites in the narrow zone of the blood tubes quality of the microscope and the staining reagents are also (Siciliano and Alano, 2015; About Malaria CDC, 2019). This limiting factors (Malaria diagnosis and treatment CDC, 2019). technique requires well-trained personnel, specialized False-negative cases lead to the unnecessary use of antibiotics, instrumentation, is costlier than conventional light microscopy, other consultations and, in some cases, progression to severe and is difficult to determine the species and number of parasites malaria. False-positive cases imply a misdiagnosis, unnecessary use (Tangpukdee et al., 2009). of anti-malaria drugs, and suffering their potential side effects (Poostchi et al., 2018). However, microscopic examination of thin and thick blood smears is commonly used in endemic areas and Rapid diagnostic tests resource-poor settings, due to its availability and easy handling. Other diagnostic techniques could complement and improve Rapid Diagnostic Tests (RDTs) are a suitable option and traditional microscopic examination and resolve its limitations. complement for detecting Plasmodium infection. RDTs are lateral-flow immunoassays that allow visualization of specific antigen–antibody recognition events (Response plan to phrp2 Quantitative Buffy Coat gene deletions WHO, 2019). They confer a qualitative diagnosis with a fast response time of less than 30 min (Cunningham et al., The Quantitative Buffy Coat (QBC) test is a qualitative 2019). RDTs depend on the observation of a visible band on a screening method for rapidly detecting the presence of malaria nitrocellulose strip produced by the capture of dye-labelled parasites in centrifuged capillary and venous blood (QBC Malaria antibodies. A drop of peripheral blood and a buffer solution are Test, 2007). Blood is centrifuged in specially coated QBC tubes usually used to perform the diagnosis on the RDT device by and visualized by optical fluorescence microscopy. The technique detecting specific Plasmodium antigens. The majority of RDTs is based on a density gradient that separates the blood cells and are based on the detection of the P. falciparum-specific protein allows the identification of parasitic forms by fluorescent histidine-rich protein II (HRP2) or universal antigen target for microscopic observation of the capillary tube. The dye commonly all malaria parasites, such as Plasmodium lactate dehydrogenase used is acridine orange, which allows the identification of parasites (p-LDH) or aldolase (Tangpukdee et al., 2009). HRP2 is between the erythrocyte and leukocyte areas. The QBC tubes also localized in the cytoplasm of P. falciparum and on the surface have an anticoagulant for the correct visualization of the sample membrane of infected erythrocytes (Murray and Bennett, 2009). and to avoid artefacts due to blood clotting (QBC Malaria Test, Gene deletions of the parasite target gene pfhrp2 are observed in 2007). QBC presents higher sensitivity and specificity than several studies in endemic areas such as Ethiopia and Bangladesh conventional thick blood smear diagnosis due to the additional (Wongsrichanalai et al., 2007; Bejon et al., 2014; Giemsa staining Frontiers in Microbiology 05 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 of malaria blood films WHO, 2016; Treatment of malaria CDC, employing Bacillus stearothermophilus DNA polymerase (Morris 2013; Nima et al. 2017). False-negative results due to pfhrp2/3 and Aydin-Schmidt, 2021). It has a 99% sensitivity and 93% gene mutation could trigger an incorrect diagnosis. Low parasite specificity for malaria parasite detection compared with density, incorrect interpretation of results, or P. malariae and microscopy and does not require thermocyclers (Ocker et al., P. ovale infections are also causes of false-negative results and 2016). A fluorescence spectrophotometer is usually needed to reasons for an incorrect diagnosis by RDTs (Kavanaugh et al., read-out diagnostic results, which restricts the applicability in 2021). A prozone effect due to excess antigen could trigger an rural areas. However, new LAMP assays are designed with a incorrect diagnosis, although it is not a common event (Gillet fluorescence readout unit in order to detect P. falciparum parasites et al., 2009). False-positive results are less common and can also (Puri et al., 2022). It is not widely implemented as a diagnostic trigger an incorrect diagnosis. Cross-reactivity due to high method, although it is postulated as an interesting alternative to parasite levels or the presence of other disease antigens are the conventional PCR methods and could be progressively main causes of false-positive results (Orish et al., 2018; implemented in resource-poor settings (Selvarajah et al., 2020). Kavanaugh et al., 2021; Pyle-Eilola et al., 2021). RDTs are a useful diagnostic support feature for conventional diagnosis, however, they cannot substitute microscopy examination Other diagnostic techniques (Ajakaye and Ibukunoluwa, 2020). Serology is based on the detection of antibodies against blood-stage malaria parasites. It is not commonly used for a rapid Polymerase chain reaction malaria diagnosis, although it is mainly used to perform seroprevalence studies of the disease. As an example, Polymerase Chain Reaction (PCR) diagnosis is a suitable Immunofluorescence Antibody Testing (IFA) uses specific alternative to conventional techniques. It is based on the antigens for the quantification of IgG and IgM antibodies in serum amplification of Plasmodium DNA, and has high sensitivity, samples (Tangpukdee et al., 2009). Combined strategies using specificity and relatively low complexity (Leski et al., 2020). It is serological, antigen detection, and DNA data are used to estimate more sensitive than microscopy and capable of identifying malaria malaria transmission and perform epidemiological studies parasites at the species level when conventional methods are not (Oviedo et al., 2020). able to detect the parasite (Johnston et al., 2006). In addition, the Flow cytometry is a laser-based cell counting method that determination of Plasmodium species by PCR assay allows the allows the quantification of erythrocytes infected by malaria unequivocal diagnosis in mixed species infection (Siwal et al., parasites. It offers automated parasite level counts and has a low 2018) or low parasite levels (Haanshuus et al., 2019), which are sensitivity (Poostchi et al., 2018). New advances based on difficult to detect by microscopic examination. Some of the main fluorescence flow cytometry have shown that the Sysmex XN-31 disadvantages of PCR diagnosis are the implementation of a device can determine the Plasmodium species and quantify non-routine technique in remote areas, the long-time (2–3 h) parasites in blood. However, it can generate false positive results needed for diagnosis, and the high cost of the technology in case of abnormal erythrocytes cell morphology and the device (Poostchi et al., 2018). Nowadays, PCR is being implemented as a was tested in a non-endemic region (Khartabil et al., 2022). diagnostic technique for malaria, although it is not the gold Biomarkers are cellular, biochemical, or molecular alterations standard procedure and is not more widely used in endemic that indicate the presence of biological, pathogenic, or therapeutic countries. Molecular techniques are useful to detect asymptomatic responses, with a high potential for diagnosis (Jain et al., 2014). patients or those with very low parasite levels; their performance The development of malaria biomarker detection, multiplex with this casuistry is considerably better than the other diagnostic biomarkers for multiple Plasmodium parasite infections, and techniques employed (Mwenda et al., 2021). This molecular biosensors are new improvements to be considered as diagnostic diagnosis technique is commonly used in high-income countries tools (Krampa et al., 2017). or regions to perform epidemiological studies (Li et al., 2014; Eshag et al., 2020; Feufack-Donfack et al., 2021). As an example, novel PCR assay, such as MC004 RT-PCR, is demonstrated to be a Diagnostic methods comparison useful tool for clinical settings and has a high degree of sensitivity and specificity (Beyene et al., 2022). To perform a comparison between the different diagnostic methods for malaria parasite detection it is important to consider the parasite levels. Low parasite levels are related to Loop-mediated isothermal lower sensitivity values due to the less number of parasites in amplification blood. Higher parasite levels are easier to detect with all the aforementioned techniques, although in some specific cases, a Loop-Mediated Isothermal Amplification (LAMP) is a prozone effect could trigger antigen detection issues by RDTs molecular technique based on the amplification of nucleic acids (Gillet et al., 2009). The commercial brand of the techniques Frontiers in Microbiology 06 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 TABLE 2 Sensitivity and specificity of malaria diagnostic methods. Diagnostic methods Sensitivity Specificity Specifications References Microscopy 75.20% 97.12% Comparative study.Thick blood films are Bejon et al. (2006), 20–40 times more sensitive than thin Wangai et al. (2011), blood films. Parasite density interferes Feleke et al. (2021) with the final result and is crucial to obtain a reliable conclusion. QBC 55.9% 88.8% Lagos State University Teaching Hospital. Adeoye and Nga (2007) 70.5% 92.1% University College Hospital, Ibadan, Oyo Ifeorah et al. (2017) State, Nigeria. RDTs 84.2% 99.8% BinaxNOW test. DiMaio et al. (2012) 63.4–100% 53.4–99.9% Mixed brands (Comparative study). Boyce and O’Meara (2017) 84.2% 95.2% University College Hospital, Ibadan, Oyo Ifeorah et al. (2017) State, Nigeria / pLDH RDT Optimal. 37–88% (37 and 51% in asympt.) 93–100% (28% in one outlier) Mixed brands (Comparative study). Feleke et al. (2021) LAMP 100% 86–99% LAMP compared with PCR. Feleke et al. (2021), Puri et al. (2022) 95–98% 91–99% LAMP compared with PCR. Morris and Aydin-Schmidt (2021) PCR Considered 100% Considered 100% Used as a reference to be compared with Feleke et al. (2021), other techniques. Mwenda et al. (2021) QBC: Quantitative Buffy Coat, RDTs: Rapid Diagnostic Tests, pLDH: Plasmodium lactose dehydrogenase, LAMP: Loop-Mediated Isothermal Amplification, PCR: Polymerase Chain Reaction, asympt: Asymptomatic. (QBC, RDTs, and PCR), RDT storage conditions, and response higher importance in medical and biomedical studies (Das et al., time are crucial for the correct interpretation of diagnostic 2015). Several software applications and tools are being developed results and could affect the final outcome. In the case of thick to detect malaria parasites in thick and thin blood smear sample and thin blood smears, the expertise of the microscopist is images using conventional light microscopy (Luengo-Oroz et al., determinant. The reference technique used as the standard 2012; Dallet et al., 2014; Das et al., 2015; Pirnstill and Coté, 2015; against which others are compared to evaluate the quality of the Bashir et al., 2017; Oliveira et al., 2017; Laketa, 2018; Manescu method is also decisive (Feleke et al., 2021). Table 2 shows the et al., 2020; Yang et al., 2020; Yu et al., 2020). diagnostic methods most commonly used for malaria parasite Deep learning is a set of computational AI processes and detection in terms of sensitivity and specificity. In some cases, methodologies that allow automated learning and the generation no differentiation between thick and thin blood smear samples of algorithms by emulating the human brain. It is based on was observed in comparative studies and meta-analyses to databases information, and uses artificial neural networks with determine sensitivity and specificity. However, thick blood multiple layers to train and generate AI algorithms (Alzubaidi smears provide a higher sensitivity than thin blood smear et al., 2021). Deep learning has, in many aspects, boosted and samples (Wangai et al., 2011). PCR is considered to have 100% improved the procedure for traditional computer vision imaging sensitivity and specificity and is usually used as the techniques (M K. Georgieff, 2016). Convolutional Neural reference method. Networks (CNN) are artificial neural networks widely used as trained classifier models to detect objects in images or videos by deep learning algorithms. Specifically, CNN classification is Novel diagnostic tools by using applied in medical diagnosis to analyse and extract efficient image analysis techniques features from images as an AI healthcare tool (Sarvamangala and Kulkarni, 2021). Imaging radiology techniques for early The global health impact of malaria has accelerated the diagnosis and treatment of emerging infectious diseases such as development and implementation of novel diagnostic strategies to Zika, Ebola, or Chikungunya are other image analysis fight against the disease. Novel diagnostic techniques based on applications (Jardon et al., 2019). Microscope image analysis image analysis and AI are being developed for malaria parasite using a U-Net (convolutional network architecture) to segment detection; an emulation of microscopic visualization by image and detect Leishmaniosis (Górriz et al., 2018) is a representative capturing and processing could be a fast and efficient alternative study of the wide variety of possibilities of CNNs. The high to performing the diagnosis. In the last years, computational computing capacity achieved over the past years and the microscopic imaging methods for object detection have held increased amount of training data for CNNs have boosted the Frontiers in Microbiology 07 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 TABLE 3 Visual image differences between thick and thin blood smear samples to distinguish malaria forms by artificial intelligence techniques. (A) Thick blood smear sample 1,000x Giemsa staining. WBC nuclei and immature trophozoites (T) are distinguished with an arrow. (B) Thin blood smear sample 1,000x Giemsa staining. Erythrocytes infected with young trophozoites (T) of P. falciparum and uninfected erythrocytes (RBC) distinguished with an arrow. Maurer dots are present in infected erythrocyte morphology. Thick blood smear sample Thin blood smear sample (A) (B) Main features Main features – Positive/Negative diagnosis. – Plasmodium species identification by parasite and erythrocyte morphology. – Possible to distinguish all development stages of the blood life cycle of the parasites. – Parasite development stages identification inside erythrocytes. – Non-species identification (except in the case of P. falciparum gametocytes). – High specificity. – Haemolysis of erythrocyte cells. – Fewer artefacts and confusion forms. – Amorphous morphology of immature Plasmodium trophozoite cytoplasm. – Fixing sample with methanol in Giemsa staining technique. – High sensitivity. – Erythrocyte and staining artefacts. – Common appearance of Giemsa artefacts. – Thick blood smear malaria parasite detection by artificial intelligence imaging tools – Thin blood smear malaria parasite detection by artificial intelligence imaging tools (Rosado et al., 2016; Xiong et al., 2019; Manescu et al., 2020; Yang et al., 2020; Yu (Ross et al., 2006; Tek et al., 2010; Dallet et al., 2014; Kareem et al., 2012; Mushabe et al., 2020). et al., 2013; Oliveira et al., 2017; Rosado et al., 2017; Sankaran et al., 2017; Dantas Oliveira et al., 2018; Pillay et al., 2019; Pardede et al., 2020; Yu et al., 2020; Abubakar et al., 2021; Davidson et al., 2021). use of this technology for medical applications (O’Mahony traditional parasite microscopy visualization, as demonstrated in et al., 2020). several studies (Xiong et al., 2019; Manescu et al., 2020; Yang et al., In particular, automated microscopy imaging analysis could 2020). Other important factors for the visualization of thick blood also be an alternative to conventional microscopy examination for smears are erythrocyte haemolysis, WBC cytoplasm rupture, and malaria diagnosis. The preparation and type of sample are the variable shapes of ring-stage trophozoites. Most of the important facts to consider to perform the correct identification of methods published for malaria parasite identification are based on biological features. Table 3 summarizes the visual image differences supervised procedures that require a previous manual labelling between thick and thin blood smears and their analysis by AI procedure of microscopic images. Malaria digital images of thick techniques (Ross et al., 2006; Mushabe et al., 2013; Dallet et al., and thin blood smears need to be labelled to create a dataset large 2014; Oliveira et al., 2017; Sankaran et al., 2017; Dantas Oliveira enough to allow the generation of an optimal detection model et al., 2018). Thick blood smear examination is crucial for a correct (Shambhu et al., 2022). This process requires to manually define diagnosis of malaria, allowing the consequent visualization of thin the bounding box of each parasite of a set of images to train the blood smears for species identification (Figure 2). Thick blood neural network model. smears are more sensitive and appropriate for low malaria parasite levels (Dowling and Shute, 1966). Nevertheless, the frequency of artefacts observed in this type of sample is higher in comparison Image acquisition with thin blood smears (Prairie, 2012). CNNs for the detection of malaria parasites in thick blood Image capturing/acquisition is the first step towards generating smears are less used in comparison with thin blood smears. New an image database for future analysis and identification. Acquisition automated parasite detection in thick blood smears based on deep depends on the equipment and infrastructure of the laboratories. learning and neural networks is an optimal alternative to Microscope-integrated cameras are often used to acquire digital Frontiers in Microbiology 08 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 images with conventional light microscopy. However, smartphone colours, cell texture, and morphology are carefully chosen cameras with an adapter bracket are an affordable alternative for (Poostchi et al., 2018). For example, erythrocyte feature automated malaria diagnosis applications (Srikanth et al., 2008; calculations in thin blood smear images are performed by open- Dallet et al., 2014; Rosado et al., 2016, 2017; Oliveira et al., 2017; source platforms such as PyRadiomics 2.2.0 (Savkare and Narote, Yu et al., 2020). Thus, in low-income countries, smartphone 2015). Feature extraction facilitates the subsequent learning and cameras would be a useful tool for acquiring digital images and classification steps by providing quantitative information on replacing integrated microscope cameras, which are usually more certain image parameters. expensive. The quality and resolution of the digital image, pixel Machine learning or pattern recognition is the final step of the morphology and density would determine future image processing image analysis procedure before identification. Classification and analysis. Other types of techniques for acquiring malaria methods are used for the identification of parasites and WBCs in parasite images with different microscopes are also used, such as thick blood smear samples, or infected and uninfected erythrocytes fluorescent microscopy, binocular microscopy, or polarized in thin blood smears. It is important to distinguish between the microscopy (Poostchi et al., 2018). Nevertheless, image acquisition parasite identification procedures for the two sample types. In both with conventional light microscopy is the most similar procedure cases, the performance of the technology developed should to emulate conventional microscopic malaria diagnosis in endemic be optimized in terms of accuracy, sensitivity, and specificity countries. Image acquisition is the first step for both traditional (Poostchi et al., 2018). Most articles published on the identification image processing techniques and deep learning methods (Hegde of malaria parasites in thick blood smears are for P. falciparum et al., 2019). infections (Yang et al., 2020). Thin blood smear parasite identification is used to distinguish between erythrocytes infected or uninfected with malaria parasites. In addition, parasite species Traditional image processing techniques identification and the development stage of the parasite in thin for malaria parasite detection blood smears are detected by using traditional pattern recognition techniques that include, for example, Support Vector Machine Image pre-processing is used in traditional computer vision (SVM) or logistic regression classifiers (Tek et al., 2010). Response techniques to automatically detect parasites and allows the time depends on the computational complexity of the predictive preparation of acquired images to improve further analysis. Most model. Complexity increases the time of response, although an studies perform noise reduction, enhancement of image contrast, evaluation between complexity and time is crucial to perform a and image resizing. These modifications would facilitate future correct and sufficiently fast identification (Freire et al., 2021). procedures of feature extraction. As an example, Gaussian average filters or low-pass filters are used to reduce the noise of malaria microscopy images (Fatima and Farid, 2020). Moreover, background Convolutional neural networks for image assumption and colour normalization and correction to malaria parasite detection reduce the effects of illumination is an affordable solution to reduce image errors (Tek et al., 2016). Colour normalization and grey Convolutional neural networks are computational systems world-based colour normalization are pre-processing methods to inspired by biological neurons designed to process data (Anwar minimise sample staining issues that could trigger image artefacts. et al., 2018). Image input is analysed to recognize visual patterns Pre-processing imaging methods for smartphone image acquisition and complete the future identification of objects as an output. by colour normalization and background removal are useful tools Neurons in deep networks are controlled by an activation to prepare images for the automated diagnosis of leishmaniasis or function, which is responsible for controlling the output. bartonellosis in remote locations (Cesario et al., 2012). Image Operations such as pooling and regularizers, with L1, L2 norms, resolution and quality are decisive to perform a correct and precise batch normalization, or dropout are key elements to make the diagnosis via imaging methods. predictive models learn better and faster (Goodfellow et al., 2016; Image segmentation is very often required to extract features. Anwar et al., 2018). Overfitting issue due to a memorization of Segmentation consists of classifying each pixel as part of the data instead of learning could interfere in the final training objects in the original image. Morphological operations, Hough outcome and obtaining of robust final predictions (Demšar and transform, K-means clustering, watershed algorithm, edge-based Zupan, 2021). segmentation algorithms, rule-based segmentation, template An important fact to consider when training CNNs is to have matching, and marker-controlled watershed are segmentation sufficient representative data. Data is commonly distributed into techniques used for thin and thick blood smear images, among three sets: training, validation, and testing. The prediction model other applications (Poostchi et al., 2018). Many of these are learns from the multiple examples of the dataset and the same complemented with thresholding techniques as a final step to training data is fed into the CNN repeatedly in an iterative extract and define the different segmented regions. procedure. During training, the validation dataset allows Feature extraction is the next procedure. The characterization hyperparameter tuning and model evaluation by a continuous of thin and thick blood smear images by features such as staining optimization. Finally, a test dataset is used to assess the model Frontiers in Microbiology 09 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 after completing the training process with unseen data (Xu and is nowadays used with promising results as DACNN model Goodacre, 2018). demonstrates (Oyewola et al., 2022). Object detection deep learning models are able to identify and CNNs have been shown to have optimum performance with locate objects of a certain class in images and videos (Jiao et al., computer-aided image diagnosis applications in specific fields of 2019). During the last few years, object detection models have study and can be generalised for other medical imaging tasks been improved and most of the state-of-the-art object detectors (Shin et al., 2016). Object detection models, such as YOLOv3, use deep learning networks. Usually, raw images need a simple YOLOv4 and YOLOv5 are used for malaria parasite detection pre-processing to resize them and fed them into the network. The (Abdurahman et al., 2021; Rocha et al., 2022). Feature scale and model itself decides and computes the appropriate features and addition of detection layers are modifications that provide better provides an output that leads to the identification and location of performances than state-of-the-art articles. Moreover, Faster objects. Among other uses, medical imaging may benefit from R-CNN (Hung and Carpenter, 2017; Ren et al., 2017) and SPPnet object detection techniques, in particular, it could be a useful (Zhou et al., 2018) are optimized neural networks used to speed alternative to malaria parasite detection (Jiao et al., 2019). up and enhance identification time. Recent studies demonstrate Object detectors are classified as two-stage or one-stage. the potential of CNNs for malaria parasite detection with Two-stage detectors have high localization and object recognition promising results, such as VGG-19 model by transfer learning accuracy, whereas one-stage detectors achieve high inference mechanism (Alnussairi and İbrahim, 2022; Jameela et al., 2022) or speed (Jiao et al., 2019). The most representative two-stage transformer-based models to obtain optimized performance detector is Faster R-CNN (Ren et al., 2017) and one-stage object parameters (Islam et al., 2022). The general procedure for malaria characteristic detectors are YOLO (Redmon et al., 2016) and SSD parasite detection using deep learning imaging methods is (Konishi et al., 2016). represented in the bottom part of Figure 3. Nowadays, CNNs have In most cases, manually labelled data is required to perform improved and replaced the use of traditional methods. all the aforementioned processes. Unsupervised training is an alternative, although most medical imaging studies are performed with supervised training data. Supervised learning based on image Automated malaria parasite level annotation is diverse and several strategies have been described calculations using deep learning (Sarangi, 2014). Whole-image classification is the annotation of methods the whole image as a type or class. Non-discerning objects are detected in the image, so the whole image is classified as a type. Automated parasitaemia counting by image analysis is a useful Object detection using bounding boxes within each image is tool that could overcome and provide support to manual parasite another option when solving classification tasks. It requires a more level quantification. Conventional malaria parasite level calculations time-consuming supervised annotation procedure of the different by microscopy visualization of thick and thin blood smears are not objects in the image. CNNs use the dataset and identify every precise and difficult to reproduce. An estimation is usually bounding box as an object class (Ibrahem et al., 2022). performed and, when parasite levels are high, is tedious and time- In the case of image segmentation, the identification of objects consuming. Thick blood smear parasite level quantification is is based on a pixel-by-level classification. Each pixel is classified as routinely performed by counting the number of parasites and a class object with its own value and annotations are manually leukocytes in a blood smear sample (WHO and Regional Office for added to images. However, it is an even higher time-consuming the Western Pacific, 2016). Thin blood smear quantification is based task for large databases, therefore automatic annotation on the counting of infected erythrocyte cells in each microscopy procedures are being developed. Thus, other conventional field (WHO and Regional Office for the Western Pacific, 2016). machine learning methods and deep learning procedures are used Quantification of parasite levels via digital image analysis techniques to automatically annotate images (Murthy et al., 2015; Cao would require the shortest period of time. An image analysis et al., 2020). software was developed to perform this function automatically with Furthermore, CNNs need large datasets with annotated data. thick blood smear images (Arco et al., 2015). For thin blood smear ImageNet is one of the largest available datasets of universal automated parasitaemia calculations there are image analysis tools images for researchers and non-commercial use (ImageNet, 2021). available to improve conventional manual counting. Determination In the particular case of malaria, a sufficiently large dataset of of malaria P. vivax parasite concentration is possible using image malaria annotated images is needed to train CNN models and processing techniques (Prasad et al., 2020). Plasmodium AutoCount perform an automated identification of parasites. Malaria thick is a digital image analysis tool to perform an automatic count of blood smears from the Institute of Electrical and Electronics parasites in Giemsa-stained thin blood smears (Ma et al., 2010). Engineers (IEEE) DataPort is an open-source image dataset Other image processing tools based on OpenCV software libraries (Malaria Thick Blood Smears | IEEE DataPort, 2021). Strikingly, were satisfactory in determining parasite levels in thin blood smear there are not many publicly available datasets of malaria thick and samples (Swain et al., 2018). Previously mentioned methods used thin blood smear images. Data augmentation techniques, to image processing techniques, such as noise reduction with filters artificially enlarge image datasets and obtain better performances, and binary transformations, to determine the presence of malaria Frontiers in Microbiology 10 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 FIGURE 3 Representation of the different procedures using Traditional Computer Vision Imaging methods or Deep Learning methods (Convolutional Neural Networks) for malaria parasite identification in thick and thin blood smear samples. parasites inside erythrocytes and perform a final parasite was image quality and adaptation to microscopy lenses. Nowadays, level calculation. smartphone cameras provide high image quality, although adaptation to the microscope is not as good as expected. Images could be disturbed by light microscopy issues, lens adaptation to Mobile phone applications for malaria smartphone cameras, or image quality downgrades related to parasite detection image focus. Microscope image auto-focus is also an issue to solve. The technology to fully automate the entire procedure of image Mobile phone applications are being developed for the focusing, image acquisition, and parasite identification by an automatic detection of malaria parasites (Cesario et al., 2012; independent device is still required. Rosado et al., 2016, 2017; Oliveira et al., 2017; Yu et al., 2020; Zhao As an example, Malaria Screener is an affordable and effective et al., 2020). Smartphone image capturing is a suitable and easy solution for automatic malaria parasite detection by a mobile alternative for the acquisition of blood smear images through the phone application (Yu et al., 2020). It combines image acquisition, microscope lens. Only an optical microscope, a mobile adapter, smear image analysis, and result visualization. It is a semi- and a conventional smartphone are needed to perform an imaging automated system based on digital images and CNN models to diagnosis. Mobile phone cameras could substitute integrated or predict the presence of malaria P. falciparum parasites in thin and external microscopy cameras and perform an optimum diagnosis thick blood smears. Other applications were developed to combine by image analysis. automatic detection of malaria parasites via an optical The integration of CNN predictive models in a smartphone magnification prototype with a smartphone device that performs software application is possible. Adapted CNN models perform image processing and analysis (Rosado et al., 2016). VGG16 the entire diagnosis in a single device. The coalescence between classification CNN, or other CNN models, were integrated into malaria automated diagnosis and smartphone software is a smartphone applications to automatically detect the presence of milestone and challenge for future implementation in worldwide malaria parasites inside erythrocytes in thin blood smear samples laboratories. Image analysis and deep learning procedures allow (Zhao et al., 2020). smartphones to be one of the best alternatives for the Gamification of the technology for the identification of implementation of automated malaria parasite detection. Even in malaria parasites in digital images is also an innovative application. resource-poor settings, smartphones are an available and relatively As an example, a web-based game where online volunteers analyse cheap option. One of the main problems of smartphone cameras thick blood smear images to detect malaria parasites was Frontiers in Microbiology 11 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 developed for the creation of an annotated image database aimed at the development of new and better diagnostic techniques (Luengo-Oroz et al., 2012). are being implemented in endemic areas (Gitta and Kilian, 2020). To sum up, smartphone applications might be the future for The increase of RDT usage in malaria-endemic areas is replacing autonomous image acquisition and analysis by AI technologies, microscopic examination of blood smears due to the lack of and a suitable alternative for malaria and Neglected Tropical resources and well-trained personnel. In addition, the biosocial Disease (NTDs) diagnosis. The possibility to integrate predictive situation of mothers and children in resource-poor regions, such models and image acquisition in a single device confers a wide as Imo State in south-eastern Nigeria, has an impact on the range of applications in the field of image analysis for diagnostics. increased appearance of complicated malaria cases (Iloh et al., 2013). A non-precise diagnosis or treatment due to the low availability of resources is a serious issue in endemic areas. Microscopy automation linked to Consequently, the implementation of new and affordable smartphone software technology diagnostic imaging techniques could help solve this problem. Smartphones are a portable and suitable alternative for Microscopy automation to move blood smear samples and malaria diagnosis via imaging techniques, which could capture focused images automatically is a challenging approach. be implemented in resource-poor settings and remote endemic Automation would solve the limitations related to the non-fully areas. They could improve and automate malaria diagnosis with autonomous diagnosis procedure performed. Image processing less need for resources and personnel. CNN models could methods allow automation of the diagnosis, although a person is be integrated inside smartphone software and an internet still needed to move the X-Y axis and issue focus of the connection would not be required. The provision of health centres microscope. A few studies have implemented automatic with mobile devices by governmental organisations and national hardware devices to solve this problem and optimise the programs against malaria would be a determinant factor for the automation of malaria diagnosis (Kaewkamnerd et al., 2012; correct implementation of this novel technology for malaria Gopakumar et al., 2018; Muthumbi et al., 2019). Microscopy diagnosis. However, due to the constraints specific to many adaptation is crucial to fully implement the aforementioned malaria endemic areas, this may be a major problem to technology in real clinical and diagnosis practices. Low-cost be addressed in the coming years by political willingness. The hardware optimization with 3D printing models to manufacture benefits of smartphones for diagnostics can be of significant value, specific parts or pieces of the microscope would be a suitable not only for malaria, but also for the diagnosis of many other option in resource-poor settings with endemic malaria. A tropical diseases or NTDs (Vasiman et al., 2019). Due to that, 3D-Printed portable robotic mobile-based microscope for the implementation in regional hospitals or small healthcare centres diagnosis of global health diseases is an example of the potential would be a challenge for future studies. New object detection of this technology (García-Villena et al., 2021). As mentioned models trained with smartphone camera images are suitable for before, some studies present the possibility of developing an malaria diagnosis deployment in resource-poor settings optical device that emulates or substitutes an optical microscope. (Abdurahman et al., 2021). An optical prototype with 1,000x magnification adapts to the Diagnostic performance studies to validate the technology are smartphone camera and avoids possible light issues (Rosado a must for the future implementation of a tool. The performance et al., 2016). Nevertheless, conventional optical microscopy evaluation should be carried out under ideal and resource-poor adaptation is the most suitable technique for image acquisition conditions to determine its effectiveness in different environments. and analysis by smartphone applications. There are many barriers to overcome in order to transition a product or technology from development to introduction and implementation. Some of the main barriers are the adjustment to Implementation of malaria digital the health and laboratory systems necessary to ensure effective microscope imaging diagnosis in adoption and implementation, demonstration of the technology’s resource-poor settings value, evaluation of operational viability, policy and regulatory requirements of government organizations, operation research to More than 90% of severe malaria produced by P. falciparum is evaluate the net effect of the technology in the field, distribution, estimated to affect young children under 5 years old in service and repair, and quality assurance and control Sub-Saharan Africa, in areas with resource-poor settings (Palamountain et al., 2012). (Schumacher and Spinelli, 2012; Gitta and Kilian, 2020). The gold The implementation of new diagnostic techniques in standard method for malaria diagnosis by the WHO is still laboratory environments has to be regulated and controlled by the microscopy, although this is dependent on laboratory resources Food and Drug Administration and WHO protocols and could result in diagnostic errors due to a lack of (Palamountain et al., 2012). The technology should be validated instrumentation, medical devices, or well-trained laboratory staff. and accepted by international and national authorities as Microscopic examination of blood smears and RDTs are the described (Mugambi et al., 2018). Most efforts to implement new techniques most used for malaria diagnosis and improvements diagnostic tools in resource-poor settings are focused on Frontiers in Microbiology 12 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 infectious diseases such as HIV, Tuberculosis, and Malaria. Deep could be implemented in low-resource settings (Zhao et al., understanding and coordination of the stakeholders involved in 2020). Automated parasite detection, parasite level calculations, the diagnostic development and implementation are milestones and faster diagnosis are some of the main advantages of image for the success of diagnostic interventions (Mugambi et al., 2018). analysis for malaria diagnosis. This technology could be used as a fast and precise tool to perform parasite level calculations (Ma et al., 2010). Overall, the use of smartphones and artificial Discussion and concluding intelligence techniques for diagnosis might help the global goal remarks of malaria eradication in the coming years. The support and enhancement of traditional microscopy-based diagnostic Epidemic malaria is very prevalent in Sub-Saharan Africa and techniques through the use of AI, the upgrading of laboratory tropical regions with low resources. It is still a global health issue infrastructures in malaria endemic areas and the improvement that should be solved by mosquito control strategies, rapid and of computer technology over the years may help to implement accurate diagnosis, and correct treatment (World Malaria Report such techniques in most remote areas. Integrating innovations WHO, 2021). Therefore, diagnosis is crucial for the eradication of into the current microscopy method would reinforce malaria the disease and to reduce mortality in prevalent regions. However, elimination (Nema et al., 2022). the recurrent problems in these environments with conventional Hardware automation is still in the process of optimization to microscopic examination due to lack of resources and complete the goal of independent predictive and mechanized experimented microscopists (Ngasala and Bushukatale, 2019), and diagnostics. Other limitations such as image quality dependence, the increasing failure of RDTs mainly due to gene mutations laboratory infrastructure requirement, local regulatory (Golassa et al., 2020), reinforces the necessity of developing new, organization permissions, or the necessity to create a standardized affordable, and accessible diagnostic methods for protocol for the final diagnosis should be addressed. Nevertheless, Plasmodium infection. several studies are improving predictive models, pre-processing Advances in image analysis and processing allow and techniques, microscope automation, and faster detections postulate the implementation of automated malaria diagnosis as (Sriporn et al., 2020; Masud et al., 2020). Artificial intelligence a new diagnostic tool. Thick and thin blood smears would be the improvements and better predictive algorithms due to computing samples analysed by the new technology. Traditional image power evolution could be an advance in terms of automatic image analysis techniques were used to automatically detect malaria diagnosis with optimized predictive results in the following years. parasites in thin and thick blood smears (Turrientes and López, In conclusion, with diagnostic techniques based on image analysis, 2016), as demonstrated in several studies (Tek et al., 2010). The the samples used are the same (thick and thin blood smear) and irruption of deep learning methodologies with CNNs has boosted the procedure of sample preparation, parasite observation and and improved the results for the identification of malaria parasites interpretation would be very similar to conventional microscopy. in comparison with traditional computer vision techniques. For In addition, it would provide technical support to health CNN models and, specifically medical image processing and professionals and help to automate the process in order to increase analysis, it is crucial to have a large image dataset to obtain reliable its efficiency. results. Unfortunately, open-source image datasets are not globally In this review, we have summarised the main advances, available and are usually used for individual CNN training. challenges, and limitations in the automation of malaria Open image availability would be a beneficial resource for the diagnosis using digital image analysis by AI tools. Smartphone scientific community. The ImageNet (ImageNet, 2021), parasite applications are a suitable option to integrate diagnosis image (Li and Zhang, 2020), and malaria thick blood smear technology into a single device and confer laboratories a new (Malaria Thick Blood Smears | IEEE DataPort, 2021) databases are tool for malaria and other disease diagnoses. New advances representative examples. Neural networks such as YOLO are used and improvements in AI would be the final milestone for the as CNN models to detect malaria parasites in blood smears optimisation and implementation of the technology worldwide. (Abdurahman et al., 2021). In conclusion, we are ever closer to developing a fast, efficient, In addition, the integration of CNN models into and optimum new diagnosis tool for malaria parasite detection smartphone software is possible. Thus, the implementation of available for laboratories located in malaria-endemic digital image analysis-based diagnostic tools in endemic areas regions worldwide. with smartphone applications could improve and automate malaria diagnosis by the emulation of the gold standard microscopy examination technique. As an example, fully Author contributions automated systems, such as the slide screening microscope EasyScan GO, evaluate their performance against WHO slide CRM, ES, and JJ-M: conceptualized and drafted the samples with promising results (Horning et al., 2021). manuscript, wrote the manuscript and designed the figures, CNN-based models are widely used as predictive models with AdO provided advice about artificial intelligence topics and the capacity to distinguish parasite forms and blood cells and designed the figures. FZ and ES provided continuous Frontiers in Microbiology 13 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 intellectual feedback about malaria diagnosis and the Base Technologies Group and the Information and Image protocols used nowadays. JJ-M, ES, FZ, ES, DL-C, SN, BB, Processing Group of Universitat Politècnica de Catalunya. Special MB, AV, ME, AA, and TP revised the manuscript and thanks to the Probitas Foundation for their support in the provided valuable feedback. ES and JJ-M edited and revised implementation of the project and the holistic support of the the overall manuscript. All authors agreed to be accountable WHO regarding digital imaging diagnosis of haemoparasites in for the content of the work. resource-poor settings. Funding Conflict of interest The project is funded by the Microbiology Department of Vall The authors declare that the research was conducted in the d’Hebron Universitary Hospital, the Cooperation Centre of the absence of any commercial or financial relationships that could Universitat Politècnica de Catalunya (CCD-UPC) and the be construed as a potential conflict of interest. Probitas Foundation. Publisher’s note Acknowledgments All claims expressed in this article are solely those of the We acknowledge the Microbiology Department of Vall authors and do not necessarily represent those of their affiliated d’Hebron Universitary Hospital, and the Vall d’Hebron- organizations, or those of the publisher, the editors and the Drassanes specialised centre in International Health, for its reviewers. Any product that may be evaluated in this article, or continuous advice, expertise, and infrastructure. The claim that may be made by its manufacturer, is not guaranteed or Computational Biology and Complex Systems Group, the Data endorsed by the publisher. References Abdurahman, F., Fante, K. A., and Aliy, M. (2021). Malaria parasite detection in Bejon, P., Andrews, L., Hunt-Cooke, A., Sanderson, F., Gilbert, S. C., and thick blood smear microscopic images using modified YOLOV3 and YOLOV4 Hill, A. V. (2014). Comparing Leishman and Giemsa staining for the assessment of models. BMC Bioinform. 22, 112–117. doi: 10.1186/s12859-021-04036-4 peripheral blood smear preparations in a malaria-endemic region in India. Malar. J. 5, 1–5. doi: 10.1186/1475-2875-5-104 Abubakar, A., Ajuji, M., and Yahya, I. U. (2021). Deepfmd: Computational analysis for malaria detection in blood-smear images using deep-learning features. Appl. Beyene, M. B., Teshome, S., Yehenew, A., Terefework, Z., Stuitje, A. R., Abebe, T., Syst. Innov. 4. doi: 10.3390/asi4040082 et al. (2022). Assessing the diagnostic performance of a novel RT-PCR fluorescence method for the detection of human plasmodium species. PLoS One 17, 1–11. doi: Adeoye, G. O., and Nga, I. C. (2007). Comparison of quantitative Buffy coat 10.1371/journal.pone.0272094 technique (QBC) with Giemsa-stained thick film (GTF) for diagnosis of malaria. Parasitol. Int. 56, 308–312. doi: 10.1016/j.parint.2007.06.007 Boyce, M. R., and O’Meara, W. P. (2017). Use of malaria RDTs in various health contexts across sub-Saharan Africa: a systematic review. BMC Public Health 17, Ajakaye, O. G., and Ibukunoluwa, M. R. (2020). Performance evaluation of a 470–415. doi: 10.1186/s12889-017-4398-1 popular malaria RDT in Nigeria compared with microscopy. J. Parasit. Dis. 44, 122–125. doi: 10.1007/s12639-019-01170-y Cao, J., Zhao, A., and Zhang, Z. (2020). Automatic image annotation method based on a convolutional neural network with threshold optimization. PLoS One 15, Alnussairi, M. H. D., and İbrahim, A. A. (2022). Malaria parasite detection using e0238956–e0238921. doi: 10.1371/journal.pone.0238956 deep learning algorithms based on (CNNs) technique. Comput. Electr. Eng. 103:108316. doi: 10.1016/J.COMPELECENG.2022.108316 CDC, CFDCP (2013). Treatment of Malaria (Guidelines for Clinicians), 1–8. Alzubaidi, L., Zhang, J., Humaidi, A. J., al-Dujaili, A., Duan, Y., al-Shamma, O., Cesario, M., Lundon, M., Luz, S., Masoodian, M., and Rogers, B. (2012). “Mobile et al. (2021). Review of deep learning: concepts, CNN architectures, challenges, support for diagnosis of communicable diseases in remote locations”, in ACM applications, future directions. J. Big Data 8:53. doi: 10.1186/s40537-021-00444-8 International Conference Proceeding Series. eds. H. Regenbrecht, C. Lutteroth and Springer International Publishing M. McGuire (New York, NY, United States: Association for Computing Machinery), Amir, A., Cheong, F. W., de Silva, J. R., Liew, J. W. K., and Lau, Y. L. (2018). 25–28. Infection and drug resistance Dovepress plasmodium knowlesi malaria: current Collins, W. E., and Jeffery, G. M. (2007). Plasmodium malariae: parasite and research perspectives. Infect. Drug Resist. 11, 1145–1155. doi: 10.2147/IDR.S148664 disease. Clin. Microbiol. Rev. 20, 579–592. doi: 10.1128/CMR.00027-07 Anwar, S. M., Majid, M., Qayyum, A., Awais, M., Alnowami, M., and Khan, M. K. Cunningham, J., Jones, S., Gatton, M. L., Barnwell, J. W., Cheng, Q., Chiodini, P. L., (2018). Medical image analysis using convolutional neural networks: a review. J. et al. (2019). A review of the WHO malaria rapid diagnostic test product testing Med. Syst. 42, 1–13. doi: 10.1007/s10916-018-1088-1 programme (2008–2018): performance, procurement and policy. Malar. J. 18:387. Arco, J. E., Górriz, J. M., Ramírez, J., Álvarez, I., and Puntonet, C. G. (2015). Digital doi: 10.1186/s12936-019-3028-z image analysis for automatic enumeration of malaria parasites using morphological Dallet, C., Kareem, S., and Kale, I. (2014). Real time blood image processing operations. Expert Syst. Appl. 42, 3041–3047. doi: 10.1016/j.eswa.2014.11.037 application for malaria diagnosis using mobile phones. Proceedings–IEEE Bartoloni, A., and Zammarchi, L. (2012). Clinical aspects of uncomplicated and International Symposium on Circuits and Systems, Piscataway, New Jersey, United severe malaria. Mediterr. J. Hematol. Infect. Dis. 4:e2012026. doi: 10.4084/ States: Institute of Electrical and Electronics Engineer, IEEE. 2405–2408. MJHID.2012.026 Dantas Oliveira, A., Carvalho, B. M., Prats, C., Espasa, M., Gomez i Prat, J., Lopez Bashir, A., Mustafa, Z., Abdelhameid, I., and Ibrahem, R. (2017). Detection and Codina, D., et al. (2018). “An automatic system for computing malaria parasite classification of malaria parasites using digital image processing. Int. Res. J. Eng. density in thin blood films” in Lecture Notes in Computer Science (including subseries Technol. (IRJET) 04, 87–89. Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 10657 LNCS. eds. M. Mendoza and S. Velastín Cham: Springer. 186–193. Bejon, P., Andrews, L., Hunt-Cooke, A., Sanderson, F., Gilbert, S. C., and Hill, A. V. S. (2006). Thick blood film examination for plasmodium falciparum Das, D. K., Mukherjee, R., and Chakraborty, C. (2015). Computational malaria has reduced sensitivity and underestimates parasite density. Malar. J. 5, 5–8. microscopic imaging for malaria parasite detection: a systematic review. J. Microsc. doi: 10.1186/1475-2875-5-104 260, 1–19. doi: 10.1111/jmi.12270 Frontiers in Microbiology 14 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 Davidson, M. S., Andradi-Brown, C., Yahiya, S., Chmielewski, J., O’Donnell, A. J., Horning, M. P., Delahunt, C. B., Bachman, C. M., Luchavez, J., Luna, C., Hu, L., Gurung, P., et al. (2021). Automated detection and staging of malaria parasites from et al. (2021). Performance of a fully-automated system on a WHO malaria cytological smears using convolutional neural networks. Biol. Imaging 1, e2–e13. microscopy evaluation slide set. Malar. J. 20, 1–11. doi: 10.1186/s12936-021-03631-3 doi: 10.1017/s2633903x21000015 Hung, J., and Carpenter, A. (2017). Applying faster R-CNN for object detection Demšar, J., and Zupan, B. (2021). Hands-on training about overfitting. PLoS on malaria images. IEEE Computer Society Conference on Computer Vision and Comput. Biol. 17, e1008671–e1008619. doi: 10.1371/journal.pcbi.1008671 Pattern Recognition Workshops, 2017-January, Piscataway, New Jersey, United States: Institute of Electrical and Electronics Engineer, IEEE. 1–7. doi:10.1109/ DiMaio, M. A., Pereira, I. T., George, T. I., and Banaei, N. (2012). Performance of CVPRW.2017.112. BinaxNOW for diagnosis of malaria in a U.S. hospital. J. Clin. Microbiol. 50, 2877–2880. doi: 10.1128/JCM.01013-12 Ibrahem, H., Salem, A., and Kang, H. (2022). S. LEOD-net: learning line-encoded bounding boxes for real-time object detection. Sensors 22:3699. doi: 10.3390/ Dowling, M. A., and Shute, G. T. (1966). A comparative study of thick and thin s22103699 blood films in the diagnosis of scanty malaria parasitaemia. Bull. World Health Organ. 34, 249–267. PMID: 5296131 Ifeorah, I. K., Brown, B. J., and Sodeinde, O. O. (2017). A comparison of rapid diagnostic testing (by plasmodium lactate dehydrogenase), and quantitative buffy Eshag, H. A., Elnzer, E., Nahied, E., Talib, M., Mussa, A., Muhajir, A. E. M. A., et al. coat technique in malaria diagnosis in children. Afr. J. Infect. Dis. 11, 31–38. doi: (2020). Molecular epidemiology of malaria parasite amongst patients in a displaced 10.21010/ajid.v11i2.5 people’s camp in Sudan. Trop. Med. Health 48, 1–7. doi: 10.1186/s41182-020-0192-3 Iloh, G. P., Chuku, A., Amadi, A. N., and Ofoedu, J. N. (2013). Proximate family Fatima, T., and Farid, M. S. (2020). Automatic detection of plasmodium parasites biosocial variables associated with severe malaria disease among under-five children from microscopic blood images. J. Parasit. Dis. 44, 69–78. doi: 10.1007/ in resource-poor setting of a rural hospital in eastern Nigeria. J. Fam. Med. Prim. s12639-019-01163-x Care 2, 256–262. doi: 10.4103/2249-4863.120739 Feufack-Donfack, L. B., Sarah-Matio, E. M., Abate, L. M., Bouopda Tuedom, A. G., ImageNet (2021). https://image-net.org/ (Accessed November 25, 2021). Ngano Bayibéki, A., Maffo Ngou, C., et al. (2021). Epidemiological and entomological studies of malaria transmission in Tibati, Adamawa region of Islam, M. R., Nahiduzzaman, M., Goni, M. O. F., Sayeed, A., Anower, M. S., Cameroon 6 years following the introduction of long-lasting insecticide nets. Ahsan, M., et al. (2022). Explainable transformer-based deep learning model for the Parasit. Vectors 14, 1–14. doi: 10.1186/s13071-021-04745-y detection of malaria parasites from blood cell images. Sensors 22:4358. doi: 10.3390/ s22124358 Feleke, D. G., Alemu, Y., and Yemanebirhane, N. (2021). Performance of rapid diagnostic tests, microscopy, loop-mediated isothermal amplification (LAMP) and Jain, P., Chakma, B., Patra, S., and Goswami, P. (2014). Potential biomarkers and PCR for malaria diagnosis in Ethiopia: a systematic review and meta-analysis. Malar. their applications for rapid and reliable detection of malaria. Bio. Med. Res. Int. 2014, J. 20, 384–311. doi: 10.1186/s12936-021-03923-8 1–20. doi: 10.1155/2014/852645 Freire, P. J., Osadchuk, Y., Spinnler, B., Napoli, A., Schairer, W., Costa, N., et al. Jameela, T., Athotha, K., Singh, N., Gunjan, V.K., and Kahali, S. (2022). “Deep (2021). Performance versus complexity study of neural network equalizers in learning and transfer learning for malaria detection,” in Computational Intelligence coherent optical systems. J. Lightwave Technol. 39, 6085–6096. doi: 10.1109/ and Neuroscience. Vol. 2022. ed. S. Roy Hindawi. JLT.2021.3096286 Jardon, M., Mohammad, S. F., Jude, C. M., and Pahwa, A. (2019). Imaging of García-Villena, J., Torres, J. E., Aguilar, C., Lin, L., Bermejo-Peláez, D., Dacal, E., emerging infectious diseases. Curr. Radiol. Rep. 7, 25–12. doi: 10.1007/ et al. (2021). 3D-printed portable robotic mobile microscope for remote diagnosis s40134-019-0338-4 of global health diseases. Electronics (Switzerland) 10, 1–16. doi: 10.3390/ Guintran, J.-O., Delacollette, C., and Trigg, P. (2006). Systems for the Early electronics10192408 Detection of Malaria Epidemics in Africa, 1–100 http://www.li.mahidol.ac.th/ Gillet, P., Mori, M., Van Esbroeck, M., Van den Ende, J., and Jacobs, J. (2009). thainatis/pdf-ebook/ebook77.pdf. Assessment of the prozone effect in malaria rapid diagnostic tests. Malar. J. 8, 1–7. Jiao, L., Zhang, F., Liu, F., Yang, S., Li, L., Feng, Z., et al. (2019). A survey of deep doi: 10.1186/1475-2875-8-271 learning-based object detection. IEEE Access 7, 128837–128868. doi: 10.1109/ Gitta, B., and Kilian, N. (2020). Diagnosis of malaria parasites plasmodium spp. ACCESS.2019.2939201 in endemic areas: current strategies for an ancient disease. BioEssays 42, e1900112– Johnston, S. P., Pieniazek, N. J., Xayavong, M. V., Slemenda, S. B., Wilkins, P. P., e1900138. doi: 10.1002/bies.201900138 and da Silva, A. J. (2006). PCR as a confirmatory technique for laboratory diagnosis Golassa, L., Messele, A., Amambua-Ngwa, A., and Swedberg, G. (2020). High of malaria. J. Clin. Microbiol. 44, 1087–1089. doi: 10.1128/JCM.44.3.1087-1089.2006 prevalence and extended deletions in plasmodium falciparum hrp2/3 genomic loci Kaewkamnerd, S., Uthaipibull, C., Intarapanich, A., Pannarut, M., Chaotheing, S., in Ethiopia. PLoS One 15, e0241807–e0241811. doi: 10.1371/journal.pone.0241807 and Tongsima, S. (2012). An automatic device for detection and classification of Gopakumar, G. P., Swetha, M., Sai Siva, G., and Sai Subrahmanyam, G. R. K. malaria parasite species in thick blood film. BMC Bioinform. 13:17. doi: (2018). Convolutional neural network-based malaria diagnosis from focus stack of 10.1186/1471-2105-13-S17-S18 blood smear images acquired using custom-built slide scanner. J. Biophotonics Kareem, S., Kale, I., and Morling, R. C. S. (2012). Automated malaria parasite 11:e201700003. doi: 10.1002/jbio.201700003 detection in thin blood films: a hybrid illumination and color constancy insensitive, Goodfellow, I., Bengio, Y., and Courville, A. (2016). Deep Learning, Vol. 2016. morphological approach. IEEE Asia-Pacific Conference on Circuits and Systems, MIT Press. Proceedings, APCCAS, Piscataway, New Jersey, United States: Institute of Electrical and Electronics Engineer, IEEE. 240–243. doi:10.1109/APCCAS.2012.6419016 Górriz, M., Aparicio, A., Raventós, B., Vilaplana, V., Sayrol, E., and López-Codina, D. (2018). “Leishmaniasis parasite segmentation and classification Kavanaugh, M. J., Azzam, S. E., and Rockabrand, D. M. (2021). Malaria rapid using deep learning” in Lecture Notes in Computer Science (including subseries diagnostic tests: literary review and recommendation for a quality assurance, quality Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). eds. F. J. control algorithm. Diagnostics 11:768. doi: 10.3390/diagnostics11050768 Perales and J. Kittler Elsevier, Science Direct 10945 LNCS, 53–62. Khartabil, T. A., de Rijke, Y. B., Koelewijn, R., van Hellemond, J. J., and Griffith, K. S., Lewis, L. S., Mali, S., and Parise, M. E. (2007). Treatment of malaria Russcher, H. (2022). Fast detection and quantification of plasmodium species in the United States: a systematic review. J. Am. Med. Assoc. 297, 2264–2277. doi: infected erythrocytes in a non-endemic region by using the Sysmex XN-31 analyzer. Malar. J. 21, 119–110. doi: 10.1186/s12936-022-04147-0 10.1001/jama.297.20.2264 Konishi, Y., Hanzawa, Y., Kawade, M., and Hashimoto, M. (2016). Fast 6D pose Haanshuus, C. G., Mørch, K., Blomberg, B., Strøm, G. E. A., Langeland, N., estimation using hierarchical pose trees. ECCV 1, 398–413. doi: Hanevik, K., et al. (2019). Assessment of malaria real-time PCR methods and 10.1007/978-3-319-46448-0 application with focus on lowlevel parasitaemia. PLoS One 14, 1–15. doi: 10.1371/ journal.pone.0218982 Krampa, F., Aniweh, Y., Awandare, G., and Kanyong, P. (2017). Recent Progress in the development of diagnostic tests for malaria. Diagnostics 7:54. doi: 10.3390/ Hegde, R. B., Prasad, K., Hebbar, H., and Singh, B. M. K. (2019). Comparison of diagnostics7030054 traditional image processing and deep learning approaches for classification of white blood cells in peripheral blood smear images. Biocybern. Biomed. Eng. 39, 382–392. Laketa, V. (2018). Microscopy in infectious disease research—imaging across doi: 10.1016/J.BBE.2019.01.005 scales. J. Mol. Biol. 430, 2612–2625. doi: 10.1016/j.jmb.2018.06.018 Heide, J., Vaughan, K. C., Sette, A., Jacobs, T., and Schulze Zur Wiesch, J. (2019). Leski, T. A., Taitt, C. R., Swaray, A. G., Bangura, U., Reynolds, N. D., Holtz, A., et al. Comprehensive review of human plasmodium falciparum-specific CD8+ T cell (2020). Use of real-time multiplex PCR, malaria rapid diagnostic test and microscopy epitopes. Front. Immunol. 10, 1–23. doi: 10.3389/fimmu.2019.00397 to investigate the prevalence of plasmodium species among febrile hospital patients in Sierra Leone. Malar. J. 19, 84–88. doi: 10.1186/s12936-020-03163-2 Heuschen, A. K., Lu, G., Razum, O., Abdul-Mumin, A., Sankoh, O., von Seidlein, L., et al. (2021). Public health-relevant consequences of the COVID-19 Li, P., Zhao, Z., Wang, Y., Xing, H., Parker, D. M., Yang, Z., et al. (2014). Nested pandemic on malaria in sub-Saharan Africa: a scoping review. Malar. J. 20, 339–316. PCR detection of malaria directly using blood filter paper samples from doi: 10.1186/s12936-021-03872-2 epidemiological surveys. Malar. J. 13, 1–6. doi: 10.1186/1475-2875-13-175 Frontiers in Microbiology 15 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 Li, S., and Zhang, Y. (2020). Microscopic Images of Parasites Species. Mendeley plasmodium falciparum provides robust geospatial estimates for malaria Data V3. doi: 10.17632/38jtn4nzs6.3 transmission in Haiti. Sci. Rep. 10, 8443–8449. doi: 10.1038/s41598-020-65419-w Luengo-Oroz, M. A., Arranz, A., and Frean, J. (2012). Crowdsourcing malaria Oyewola, D. O., Danda, E., Misra, S., and Damasevicius, R. (2022). A novel data parasite quantification: an online game for analyzing images of infected thick blood augmentation convolutional neural network for detecting malaria parasite in blood smears. J. Med. Internet Res. 14, e167–e113. doi: 10.2196/jmir.2338 smear images. Appl. Artif. Intell. 36. doi: 10.1080/08839514.2022.2033473 Georgieff, M. K. (2016). HHS public access. Physiol. Behav. 176, 100–106. doi: Palamountain, K. M., Baker, J., Cowan, E. P., Essajee, S., Mazzola, L. T., 10.1021/acs.molpharmaceut.7b00578.Comparison Metzler, M., et al. (2012). Perspectives on introduction and implementation of new point-of-care diagnostic tests. J. Infect. Dis. 205, S181–S190. doi: 10.1093/infdis/ Ma, C., Harrison, P., Wang, L., and Coppel, R. L. (2010). Automated estimation of jis203 parasitaemia of plasmodium yoelii-infected mice by digital image analysis of Giemsa-stained thin blood smears. Malar. J. 9:348. doi: 10.1186/1475-2875-9-348 Pardede, J., Dewi, I., Fadilah, R., and Triyani, Y. (2020). Automated malaria diagnosis using object detection retina-net based on thin blood smear image. J. Malaria Thick Blood Smears: IEEE Data Port (2021). Available at: https://ieee- Theor. Appl. Inf. Technol. 98, 757–767. dataport.org/documents/malaria-thick-blood-smears (Accessed November 15, 2021) Phillips, M. A., Burrows, J. N., Manyando, C., van Huijsduijnen, R. H., van Voorhis, W. C., and Wells, T. N. C. (2017). Malaria. Nat. Rev. Dis. Primers 3, 1–24. Manescu, P., Shaw, M. J., Elmi, M., Neary-Zajiczek, L., Claveau, R., Pawar, V., et al. doi: 10.1038/nrdp.2017.50 (2020). Expert-level automated malaria diagnosis on routine blood films with deep neural networks. Am. J. Hematol. 95, 883–891. doi: 10.1002/ajh.25827 Pillay, E., Khodaiji, S., Bezuidenhout, B. C., Litshie, M., and Coetzer, T. L. (2019). Evaluation of automated malaria diagnosis using the Sysmex XN-30 analyser in a Masud, M., Alhumyani, H., Alshamrani, S. S., Cheikhrouhou, O., Ibrahim, S., clinical setting. Malar. J. 18, 15–14. doi: 10.1186/s12936-019-2655-8 Muhammad, G., et al. (2020). Leveraging deep learning techniques for malaria parasite detection using Mobile application. Wirel. Commun. Mob. Comput. 2020, Pirnstill, C. W., and Coté, G. L. (2015). Malaria Diagnosis Using a Mobile Phone 1–15. doi: 10.1155/2020/8895429 Polarized Microscope. Sci. Rep. 5, 1–13. doi: 10.1038/srep13368 Molyneux, M. E. (1989). Malaria—clinical features in children. J. R. Soc. Med. 82 Poostchi, M., Silamut, K., Maude, R. J., Jaeger, S., and Thoma, G. (2018). Image Suppl 1, 35–38. analysis and machine learning for detecting malaria. Transl. Res. 194, 36–55. doi: 10.1016/j.trsl.2017.12.004 Morris, U., and Aydin-Schmidt, B. (2021). Performance and application of commercially available loop-mediated isothermal amplification (LAMP) kits in Prairie, K. (2012). Methods Manual, vol. 1, 1–109. malaria endemic and non-endemic settings. Diagnostics 11, 1–18. doi: 10.3390/ Prasad, N., Swanline, F., Nithyashree, N., and Kumar, M. (2020). Determination diagnostics11020336 of malaria parasite concentration using image processing. Int. J. Eng. Tech. Res. V9, Mugambi, M. L., Peter, T., F Martins, S., and Giachetti, C. (2018). How to 412–415. doi: 10.17577/ijertv9is020062 implement new diagnostic products in low-resource settings: an end-to-end Prevention CDC (2019). CDC-Malaria-Diagnosis and Treatment (United States)– framework. BMJ Glob. Health 3, e000914–e000910. doi: 10.1136/bmjgh-2018-000914 Diagnosis (U.S.). Centres for Diseases Control and Prevention CDC, CDC. Available Murray, C. K., and Bennett, J. W. (2009). Rapid diagnosis of malaria. Interdiscip. at: https://www.cdc.gov/malaria/diagnosis_treatment/clinicians1.html Perspect. Infect. Dis. 2009, 1–7. doi: 10.1155/2009/415953 Puri, M., Kaur Brar, H., Madan, E., Srinivasan, R., Rawat, K., Gorthi, S. S., et al. Murthy, V. N., Maji, S., and Manmatha, R. (2015). “Automatic image annotation (2022). Rapid diagnosis of Plasmodium falciparum malaria using a pointof- care using deep learning representations”, in ICMR 2015- Proceedings of the 2015 ACM loop-mediated isothermal amplification device. Front. Cell. Infect. Microbiol. International Conference on Multimedia Retrieval. 603–606. 12:961832. doi: 10.3389/fcimb.2022.961832 Mushabe, M. C., Dendere, R., and Douglas, T. S. (2013). Automated detection of Pyle-Eilola, A. L., Kahwash, S., Dennis, G., Hubbard, N., and Leber, A. (2021). The malaria in Giemsa-stained thin blood smears. Proceedings of the Annual brief case: a 6-year-old with fever, abdominal pain, and recent travel to Sierra Leone. International Conference of the IEEE Engineering in Medicine and Biology J. Clin. Microbiol. 3, 9–12. doi: 10.1128/JCM.00535-21 Society, EMBS, Piscataway, New Jersey, United States: Institute of Electrical and Redmon, J., Divvala, S., Girshick, R., and Farhadi, A. (2016). “You only look once: Electronics Engineer, IEEE. 3698–3701. doi:10.1109/EMBC.2013.6610346 unified, real-time object detection”, Proceedings of the IEEE Computer Society Muthumbi, A., Chaware, A., Kim, K., Zhou, K. C., Konda, P. C., Chen, R., et al. Conference on Computer Vision and Pattern Recognition, 2016-December, (2019). Learned sensing: jointly optimized microscope hardware for accurate image Piscataway, New Jersey, United States: Institute of Electrical and Electronics classification. Biomed. Opt. Express 10, 6351–6369. doi: 10.1364/boe.10.006351 Engineer, IEEE. 779–788. Mwenda, M. C., Fola, A. A., Ciubotariu, I. I., Mulube, C., Mambwe, B., Kasaro, R., Ren, M. (2019). Greater political commitment needed to eliminate malaria. Infect. et al. (2021). Performance evaluation of RDT, light microscopy, and PET-PCR for Dis. Poverty 8, 28–25. doi: 10.1186/s40249-019-0542-8 detecting plasmodium falciparum malaria infections in the 2018 Zambia National Ren, S., He, K., Girshick, R., and Sun, J. (2017). Faster R-CNN: towards real-time Malaria Indicator Survey. Malar. J. 20, 386–310. doi: 10.1186/s12936-021-03917-6 object detection with region proposal networks. IEEE Trans. Pattern Anal. Mach. Nema, S., Rahi, M., Sharma, A., and Bharti, P. K. (2022). Strengthening malaria Intell. 39, 1137–1149. doi: 10.1109/TPAMI.2016.2577031 microscopy using artificial intelligence-based approaches in India. Comment Ricci, F. (2012). Social implications of malaria and their relationships with 5:100054. doi: 10.1016/j.lansea.2022.100054, (The Lancet Regional Health- poverty. Mediterr. J. Hematol. Infect. Dis. 4:e2012048. doi: 10.4084/MJHID.2012.048 Southeast Asia) Rocha, M., Claro, M., Neto, L., Aires, K., Machado, V., and Veras, R. (2022). Ngasala, B., and Bushukatale, S. (2019). Evaluation of malaria microscopy Malaria parasites detection and identification using object detectors based on deep diagnostic performance at private health facilities in Tanzania. Malar. J. 18, 1–7. doi: neural networks: A wide comparative analysis. Comput. Methods Biomech. Biomed. 10.1186/s12936-019-2998-1 Eng. Imaging Vis. doi: 10.1080/21681163.2022.2111715 Nima, M. K., Hougard, T., Hossain, M., Kibria, M., Mohon, A., Johora, F., et al. Rosado, L., Correia da Costa, J., Elias, D., and Cardoso, J. (2016). Automated (2017). Case report: a case of plasmodium falciparum hrp2 and hrp3 gene mutation detection of malaria parasites on thick blood smears via Mobile devices. Procedia in Bangladesh. Am. J. Trop. Med. Hyg. 97, 1155–1158. doi: 10.4269/ajtmh.16-0884 Comput. Sci. 90, 138–144. doi: 10.1016/j.procs.2016.07.024 O’Mahony, N., Campbell, S., Carvalho, A., Harapanahalli, S., Hernandez, G., Rosado, L., da Costa, J. M. C., Elias, D., and Cardoso, J. S. (2017). Mobile-based Krpalkova, L., et al. (2020). “Deep learning vs. traditional computer vision” in analysis of malaria-infected thin blood smears: automated species and life cycle Advances in Intelligent Systems and Computing. vol. 943 Springer: SpringerLink, stage determination. Sensors (Switzerland) 17, 1–22. doi: 10.3390/s17102167 128–144. Ross, N. E., Pritchard, C. J., Rubin, D. M., and Dusé, A. G. (2006). Automated Ocker, R., Prompunjai, Y., Chutipongvivate, S., and Karanis, P. (2016). Malaria image processing method for the diagnosis and classification of malaria on thin diagnosis by loop-mediated isothermal amplification (LAMP) in Thailand. Rev. Inst. blood smears. Med. Biol. Eng. Comput. 44, 427–436. doi: 10.1007/s11517-006-0044-2 Med. Trop. São Paulo 58, 2–7. doi: 10.1590/S1678-9946201658027 Sarangi, N. (2014). Image annotation using deep learning techniques. 2017 Oliveira, A. D., Prats, C., Espasa, M., Zarzuela Serrat, F., Montañola Sales, C., International Conference on Intelligent Computing and Control (I2C2’17), Piscataway, Silgado, A., et al. (2017). The malaria system microapp: a new, mobile device-based New Jersey, United States: Institute of Electrical and Electronics Engineer, IEEE. tool for malaria diagnosis. JMIR Res. Protoc. 6, 1–12. doi: 10.2196/resprot. 1–20 6758 Sarvamangala, D. R., and Kulkarni, R. V. (2021). Convolutional neural networks Orish, V. N., De-Gaulle, V. F., and Sanyaolu, A. O. (2018). Interpreting rapid in medical image understanding: a survey. Evol. Intel. 15, 1–22. doi: 10.1007/ diagnostic test (RDT) for plasmodium falciparum 11 medical and health sciences s12065-020-00540-3 1108 medical microbiology. BMC. Res. Notes 11, 1–6. doi: 10.1186/ Savkare, S. S., and Narote, S. P. (2015). Automated system for malaria parasite s13104-018-3967-4 identification. Proceedings−2015 International Conference on Communication, Oviedo, A., Knipes, A., Worrell, C., Fox, L. A. M., Desir, L., Fayette, C., et al. Information and Computing Technology, ICCICT, vol. 20, 15–18. doi:10.1109/ (2020). Combination of serological, antigen detection, and DNA data for ICCICT.2015.7045660 Frontiers in Microbiology 16 frontiersin.org Maturana et al. 10.3389/fmicb.2022.1006659 Schumacher, R. F., and Spinelli, E. (2012). Malaria in children. Mediterr. J. QBC (2007). QBC ™ Malaria Test. Hematol. Infect. Dis. 4:e2012073. doi: 10.4084/MJHID.2012.073 Vasiman, A., Stothard, J. R., and Bogoch, I. I. (2019). Mobile phone devices and Selvarajah, D., Naing, C., Htet, N. H., and Mak, J. W. (2020). Loop-mediated handheld microscopes as diagnostic platforms for malaria and neglected tropical isothermal amplification (LAMP) test for diagnosis of uncomplicated malaria in diseases (NTDs) in low-resource settings: a systematic review, historical perspective endemic areas: a meta-analysis of diagnostic test accuracy. Malar. J. 19, 1–10. doi: and future outlook. Adv. Parasitol. 103, 151–173. doi: 10.1016/BS.APAR.2018. 10.1186/s12936-020-03283-9 09.001 Shambhu, S., Koundal, D., das, P., Hoang, V. T., Tran-Trung, K., and Turabieh, H. Vila, J., Gómez, M. D., Salavert, M., and Bosch, J. (2017). Métodos de diagnóstico (2022). Computational methods for automated analysis of malaria parasite using rápido en microbiología clínica: necesidades clínicas. Enferm. Infecc. Microbiol. Clin. blood smear images: recent advances. Comput. Intell. Neurosci. 2022, 1–18. doi: 35, 41–46. doi: 10.1016/j.eimc.2016.11.004, 10.1155/2022/3626726 Wangai, L. N., Karau, M. G., Njiruh, P. N., Sabah, O., Kimani, F. T., Magoma, G., Shin, H. C., Roth, H. R., Gao, M., Lu, L., Xu, Z., Nogues, I., et al. (2016). Deep et al. (2011). Sensitivity of microscopy compared to molecular diagnosis of P. convolutional neural networks for computer-aided detection: CNN architectures, falciparum: implications on malaria treatment in epidemic areas in Kenya. Afr. J. dataset characteristics and transfer learning. IEEE Trans. Med. Imaging 35, Infect. Dis. 5, 1–6. doi: 10.4314/ajid.v5i1.66504 1285–1298. doi: 10.1109/TMI.2016.2528162 WHO (2016). ‘Giemsa staining of malaria blood films’, malaria microscopy Shujatullah, F., Malik, A., Khan, H. M., and Malik, A. (2006). Comparison of standard operating procedure-MM-SOP-07A, 1–6. http://www.wpro.who.int/mvp/ different diagnostic techniques in plasmodium falciparum cerebral malaria. J. Vector lab_quality/2096_oms_gmp_sop_07a_rev.pdf Borne Dis. 43, 186–190. PMID: 17175704 WHO (2019). ‘Response plan to pfhrp 2 gene deletions’, who, 1–33 Siciliano, G., and Alano, P. (2015). Enlightening the malaria parasite life cycle: Bioluminescent Plasmodium in fundamental and applied research. Front. Microbiol. WHO (2021). WHO recommends groundbreaking malaria vaccine for children 6, 1–8. doi: 10.3389/fmicb.2015.00391 at risk. Available at: https://www.who.int/news/item/06-10-2021-who-recommends- groundbreaking-malaria-vaccine-for-children-at-risk (Accessed October Simba, D., Kakoko, D., Nyamhanga, T., Mrango, Z., and Mujinja, P. (2018). 19, 2021) Improving prompt access to malaria diagnostics and treatment in rural remote areas using financial benefit for community health workers in Kilosa district, Tanzania. Wongsrichanalai, C., Barcus, M., Muth, S., Sutamihardja, A., and Wernsdorfer, W. Res. Rep. Trop. Med. 9, 137–146. doi: 10.2147/rrtm.s172944 (2007). A review of malaria diagnostic tools: microscopy and rapid diagnostic test (RDT). Am. J. Trop. Med. Hyg. 77, 119–127. doi: 10.4269/ajtmh.2007.77.119 Siwal, N., Singh, U. S., Dash, M., Kar, S., Rani, S., Rawal, C., et al. (2018). Malaria diagnosis by PCR revealed differential distribution of mono and mixed species World Health Organization and Regional Office for the Western Pacific (2016). infections by plasmodium falciparum and p. vivax in India. PLoS One 13, 1–14. doi: Malaria parasite counting. Malaria Microscopy Standard Operating Procedure, 1–5. 10.1371/journal.pone.0193046 Available at: https://apps.who.int/iris/handle/10665/274382 Srikanth, P., Ezhil, R., Suchitra, S., Anandhi, I., Maheswari, U., and Kalyani, J. World Malaria Report (2021). World heal (2021). Geneva Licence: CC BY-NC-SA (2008). The Mobile phone in a tropical setting-emerging threat for infection control. 3.0 IGO Int. J. Infect. Dis. 12:e367. doi: 10.1016/j.ijid.2008.05.973 Xiong, Y., Du, B., and Yan, P. (2019). Reinforced Transformer for Medical, vol. 1, Sriporn, K., Tsai, C. F., Tsai, C. E., and Wang, P. (2020). Analyzing lung disease 673–680 doi: 10.1007/978-3-030-32692-0. using highly effective deep learning techniques. Healthc. 8, 1–21. doi: 10.3390/ Xu, Y., and Goodacre, R. (2018). On splitting training and validation set: A healthcare8020107 comparative study of cross-validation, bootstrap and systematic sampling for Swain, M., Dhariwal, S., and Kumar, G. (2018). A python (open CV) based estimating the generalization performance of supervised learning. J. Anal. Testing 2, automatic tool for parasitemia calculation in peripheral blood smear. Proceedings 249–262. doi: 10.1007/s41664-018-0068-2 -2nd International Conference on Intelligent Circuits and Systems, ICICS 2018, Yang, F., Poostchi, M., Yu, H., Zhou, Z., Silamut, K., Yu, J., et al. (2020). Deep Piscataway, New Jersey, United States: Institute of Electrical and Electronics learning for smartphone-based malaria parasite detection in thick blood smears. Engineer, IEEE. 445–448. doi:10.1109/ICICS.2018.00096 IEEE J. Biomed. Health Inform. 24, 1427–1438. doi: 10.1109/JBHI.2019. Talapko, J., Skrlec, I., Alebic, T., Jukic, M., and Vcev, A. (2019). Malaria: the past 2939121 and the present. Microorganisms 7, 1–17. doi: 10.3390/microorganisms7060179 Yu, H., Yang, F., Rajaraman, S., Ersoy, I., Moallem, G., and Poostchi, M. (2020). Tangpukdee, N., Duangdee, C., Wilairatana, P., and Krudsood, S. (2009). Malaria screener: a smartphone application for automated malaria screening. BMC Malaria diagnosis: A brief review. Korean J. Parasitol. 47, 93–102. doi: 10.3347/ Infect. Dis. 20, 825–828. doi: 10.1186/s12879-020-05453-1 kjp.2009.47.2.93 Zekar, L., and Sharman, T. (2021). Plasmodium Falciparum Malaria. ed. C. G. Tek, F. B., Dempster, A. G., and Kale, I. (2010). Parasite detection and identification Casey Stat Pearls. for automated thin blood film malaria diagnosis. Comput. Vis. Image Underst. 114, Zhao, O. S., Kolluri, N., Anand, A., Chu, N., Bhavaraju, R., Ojha, A., et al. (2020). 21–32. doi: 10.1016/j.cviu.2009.08.003 Convolutional neural networks to automate the screening of malaria in low-resource Tek, F. B., Dempster, A. G., and Kale, İ. (2016). Adaptive gray world-based color countries. PeerJ 8, e9674–e9620. doi: 10.7717/peerj.9674 normalization of thin blood film images. Available at: http://arxiv.org/abs/1607.04032 Zhou, H., Wang, Y., and Ye, M. (2018). “A method of CNN traffic classification Turrientes, M., and López, R. (2016). Diagnóstico De Laboratorio Y Profilaxis De based on sppnet”, in Proceedings -14th International Conference on Computational La Malaria. Control Calidad SEIMC, 1–5. Available at: https://seimc.org/contenidos/ Intelligence and Security, CIS 2018, Piscataway, New Jersey, United States: Institute ccs/revisionestematicas/parasitologia/malaria.pdf of Electrical and Electronics Engineer, IEEE. 390–394. Frontiers in Microbiology 17 frontiersin.org