diagnostics Technical Note End to End Colonic Content Assessment: ColonMetry Application Bernat Orellana 1 , Eva Monclús 1 , Isabel Navazo 1 , Álvaro Bendezú 2 , Carolina Malagelada 3,4,5 and Fernando Azpiroz 3,4,5, * 1 Visualization, Virtual Reality and Graphics Interaction Research Group, UPC-BarcelonaTech, 08034 Barcelona, Spain 2 Digestive Department, University Hospital General de Catalunya, 08190 Barcelona, Spain 3 Digestive System Research Unit, University Hospital Vall d’Hebron, 08035 Barcelona, Spain 4 Departament de Medicina, Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain 5 Centro de Investigación Biomédica en Red de Enfermedades Hepáticas y Digestivas, Ciberehd, 28029 Madrid, Spain * Correspondence: Fernando.Azpiroz@uab.cat; Tel.: +34-932746259 Abstract: The analysis of colonic contents is a valuable tool for the gastroenterologist and has multiple applications in clinical routine. When considering magnetic resonance imaging (MRI) modalities, T2 weighted images are capable of segmenting the colonic lumen, whereas fecal and gas contents can only be distinguished in T1 weighted images. In this paper, we present an end-to-end quasi-automatic framework that comprises all the steps needed to accurately segment the colon in T2 and T1 images and to extract colonic content and morphology data to provide the quantification of colonic content and morphology data. As a consequence, physicians have gained new insights into the effects of diets and the mechanisms of abdominal distension. Keywords: magnetic resonance imaging; colonic content; colon segmentation; intestinal biomass; intestinal gas; functional digestive disorders Citation: Orellana, B.; Monclús, E.; 1. Introduction Navazo, I.; Bendezú, Á.; Malagelada, About half of the patients in gastroenterological practice complain of digestive symp- C.; Azpiroz, F. End to End Colonic toms without detectable abnormalities by conventional diagnostic tests. In a large propor- Content Assessment: ColonMetry tion of these patients, specifically those with intestinal functional disorders, their symptoms Application. Diagnostics 2023, 13, 910. seem related to colonic dysfunction and abnormal intraluminal contents; however, the eval- https://doi.org/10.3390/ uation of the volume and distribution of solid and gas content in the colon is fraught with diagnostics13050910 methodological constrains. Abdominal imaging by means of computer tomography (CT) Academic Editor: Giuseppe has shed some light into the physiology of colonic handling of contents and disease [1–5]. Ingravallo Magnetic resonance imaging (MRI) has recently become an alternative technique to CT, since it provides an appropriate visualization of soft tissues while avoiding the exposure Received: 15 December 2022 to ionizing radiation used in CT [6]. The MRI acquisition technique allows the modifica- Revised: 17 February 2023 Accepted: 24 February 2023 tion of the signal emitted to acquire the different tissues presented in the body. The MRI Published: 28 February 2023 modalities commonly used for the analysis of colonic content are T2-weighted (T2) and T1-weighted fat-SAT (T1-FS). However, those techniques, by themselves, do not allow the measurement of the colonic content. T2 sequence offers an optimal visualization of the colon boundaries, contrasting the colon lumen (dark voxels) against the fat around the Copyright: © 2023 by the authors. colon (bright voxels) (Figure 1a). T1-FS sequence (Figure 1b) provides visual discrimination Licensee MDPI, Basel, Switzerland. of the colonic content: feces are bright regions, while colonic gas is dark. Therefore, the This article is an open access article information found in T1-FS and T2 images is complementary and both modalities are distributed under the terms and required to obtain a quantitative assessment of colonic morphology and content. To study conditions of the Creative Commons the gut in normal conditions, it is necessary to acquire the abdominal images without Attribution (CC BY) license (https:// preparation, e.g., administration of drugs or contrast. creativecommons.org/licenses/by/ 4.0/). Diagnostics 2023, 13, 910. https://doi.org/10.3390/diagnostics13050910 https://www.mdpi.com/journal/diagnostics Diagnostics 2023, 13, x FOR PEER REVIEW 2 of 10 both modalities are required to obtain a quantitative assessment of colonic morphology Diagnostics 2023, 13, 910 and content. To study the gut in normal conditions, it is necessary to acquire the 2 ofab- 10 dominal images without preparation, e.g., administration of drugs or contrast. (a) (b) Figure 1. Figure 1. (a) (a) MRI MRI T2 T2 coronal coronal slice slice showing showing ascending ascending colon colon and and descending descending colon colon (green (green arrows). arrows). (b) The same area in the T1-FS modality. Arrows are colorized depending on the contents: blue (b) The same area in the T1-FS modality. Arrows are colorized depending on the contents: blue arrows point to gas and red to feces. Courtesy of [7]. arrows point to gas and red to feces. Courtesy of [7]. In general, In general, MRIMRI scans scansareareharder hardertotosegment segmentthan thanCT CTscan scanimages images duedueto to thethe lack of lack fixed correspondence between tissues and intensities and the incidence of fixed correspondence between tissues and intensities and the incidence of noise and of noise and arti- facts on on artifacts thethe images images [8].[8]. T2T2modality modality displays displaysthe thecolon colonasasaadarkdarkregion region contrasted contrasted withwith bright surrounding fat, but its appearance may suffer large intensity bright surrounding fat, but its appearance may suffer large intensity and texture variations and texture varia- tions along along its tract, its tract, mainly mainly caused caused by thebypresence the presence of colonic of colonic content. content. Regarding Regarding T1-FS, T1-FS, this this modality allows for the proper discrimination of fecal content: modality allows for the proper discrimination of fecal content: feces are bright regions feces are bright regions while colonic while colonic gasgas isis dark. dark. However, However, fat fat tissues tissues are are also also dark, dark, almost almost indistinguishable indistinguishable from the from the adjacent adjacent colon colon gas. gas. In In consequence, consequence, colonic colonic boundaries boundaries in T1-FS are in T1-FS are not not visible visible in regions in regions with with little little presence presence of of faces. faces. Since Since clinical clinical research research involves involves numerous numerous studies studies to be analyzed, and current colon colon segmentation segmentation algorithms are time-consuming for the specialists, the clinical practice would benefit from a higher degree of automation of colon segmentation algorithms. Previous studies studies evidenced evidencedthe thepractical practicallimitations limitations ofof manual manual segmentation segmentation of intes- of in- tinal imaging testinal imaging obtained obtained by different by differentMRIMRI modalities modalities([9–11]). [9–11].In this regard, In this regard,by by a close in- a close teraction between interaction between physicians physicians andandcomputer computer scientists, our group scientists, our groupdeveloped developedoriginal tools original devoted tools to colon devoted segmentation to colon segmentationin T2-weighted half-Fourier in T2-weighted acquisition half-Fourier single-shot acquisition turbo single-shot turbo spin(HASTE) spin echo echo (HASTE) (T2) and (T2) and T1-weighted T1-weighted VIBE Fat-Sat VIBE Fat-Sat (T1-FS)(T1-FS) sequences.sequences. CeballosCebal-et al. los et al. [12] developed an end-to-end application for colonic content [12] developed an end-to-end application for colonic content assessment from T2 and T1- assessment from T2 and T1-FS sequences, FS sequences, allowing allowing the quantification the quantification of theof the colon colon volume volume fromfrom T2 images T2 images andand the the fecalfecal volume volume content content fromfrom T1-FS T1-FS images. images. Moreover,some Moreover, someshape shapeproperties properties of of the the colon were were also also computed computed (perimeter (perimeter and and length). length). The application application works works on a 3D volume model built from the MRI images. Our application provided a seed region growing mechanism combined combined with the ability to add stop markers to prevent leaking. The colon segmentation obtained obtained in in T2 T2 images images was was transferred transferred to to T1-FS T1-FS images images using using aa non-rigid non-rigid semi-automatic semi-automatic registration registration method; however, a manual specialist validation evidenced that method; however, a manual specialist validation evidenced that only only the fecal the fecal content was properly identified. Once both modalities have been segmented, an automatic classification method determined the amount of fecal content inside the colon. This pipeline has been used in different clinical experiments [5,13], but it requires a relatively long time Diagnostics 2023, 13, 910 3 of 10 that the specialist had to dedicate (40 min by scans (T2 and T1-FS)), so there was space to improve basically in the two segmentation processes. Following this line, we developed a quasi-automatic segmentation algorithm for the colon on T2 images based on an unsupervised clustering algorithm guided by a probabilistic measure extracted from a training set [7,14]. The process drastically reduced the specialist interaction effort compared to other state-of the-art solutions (5 min by scan). Focusing on the automation of colon segmentation in T1-FS, we subsequently developed a specialized registration algorithm that provides an accurate and fast colon segmentation in T1-FS using as an input the colon segmentation from T2 images ([15]: under review process). From the first pipeline, many improvements have been developed to facilitate obtaining adequate colonic segmentation. However, it is important to note that the specialist always needs to supervise and validate the results after T2 and T1-FS colon segmentations have been computed, and this validation is not straightforward. For some areas, the analysis of a single modality may be confusing, and the validation may require the simultaneous visualization of both modalities. For instance, the cecum and sigma segments are very difficult to distinguish from the small bowel in T2 modality, but the detection (and validation process) is substantially simplified by simultaneous inspection of T1-FS and T2 images. To address this medical problem, the objective of the present study was to develop a complete end- to-end quasi-automatic framework to segment the colon in T2 and T1-FS MRI sequences for quantitative evaluation of colonic morphology and the discrimination between the gas and non-gaseous (solid) colonic content. In this paper, we present a novel interactive module to visualize both segmentations synchronously, to facilitate the validation of the inter-modality coherence between both segmentations, and its evaluation by domain experts. Finally, although some modules have been previously validated, here we present a complete medical validation and discussion about the integrated segmentation and interaction techniques for all the components of the pipeline. 2. Materials and Methods The application’s framework is shown in Figure 2. The patient is scanned twice to obtain T1-FS and T2-weighted images. Once the images are captured, the specialist provides a minimal set of 5 anatomical reference points along the colon in T2 images. From this information, the system automatically computes the colon segmentation in T2 images. If the obtained segmentation is not accurate enough, the specialist can add refinement markers with the objective to automatically correct misclassified regions (see Section 2.1). Next, the colon segmentation in T1-FS images is carried out using an automatic approach that uses the colon segmentation previously calculated in T2 images (see Section 2.2). Then, the medical expert can interactively visualize both segmentations to validate them (see Section 2.3). Finally, the non-gaseous and the gaseous colonic content and different properties of the colonic morphology are quantified using both segmentations (see Section 2.4). Each module of the system is able to provide a 3D visualization of the colon, which helps to understand the colon anatomy. 2.1. Colon Segmentation in T2 Images In Ref. [7] we presented a novel accurate colon segmentation algorithm from unpre- pared MRI in T2 images with a low user effort. The process only required the specialist to provide a minimal set of five anatomical reference points (marker points) along the colonic trajectory to guide the segmentation algorithm. These points determined the start and end of each of the colon segments: ascending, transverse, descending, and pelvic. Depending on the complexity of the colonic shape, more points may be required to ensure a correct segmentation result. From this information, the algorithm automatically computed the colonic segmentation. The algorithm is founded on the estimation of the colon medial path and the usage of a probabilistic model for tissue classification based on the MRI intensities and the colon medial path distance. A probability model guides a progressive segmentation Diagnostics 2023, 13, 910 4 of 10 Diagnostics 2023, 13, x FOR PEER REVIEW process 4 of 10 based on a multigrid architecture using a 3D graph-cuts algorithm. This multigrid architecture provides computational scalability to the solution. Figure 2. Figure 2. Overview Overview of of the the system: system: the the left left column column shows shows the the algorithm’s algorithm’s inputs: inputs: T1-FS T1-FS images images and and T2 images. First, the specialist provides a minimal set of 5 anatomical reference points T2 images. First, the specialist provides a minimal set of 5 anatomical reference points along the colon along the colon path. From this information, the system automatically computes the colon segmentation in T2 path. From this information, the system automatically computes the colon segmentation in T2 images. images. If the obtained segmentation is not accurate enough, the specialist can add refinement mark- If the obtained segmentation is not accurate enough, the specialist can add refinement markers with ers with the objective to automatically correct misclassified regions. Once the colon segmentation the hasobjective to automatically been validated correct by the medical misclassified expert, regions. the system Once thecomputes automatically colon segmentation has been the colon segmenta- validated by the medical expert, the system automatically computes the colon tion in T1-FS. Then, the new module allows visualizing both segmentations simultaneously segmentation in T1-FS. and Then, the new module synchronously, allows to validate thevisualizing inter-modalityboth coherence segmentations simultaneously between andthe them. Finally, synchronously, system auto- matically to validatecomputes different properties the inter-modality of the colon coherence between shape them. and the Finally, colonic content system information. automatically computes different properties of the colon shape and colonic content information. 2.1. Colon Segmentation in T2 Images 2.2. Colon Segmentation in T1-FS Images In Ref. [7] we presented a novel accurate colon segmentation algorithm from unpre- paredInMRI Ref. [15] in T2we presented images withan automatic a low colonThe user effort. segmentation process only algorithm requiredfor theT1-FS images. specialist to The algorithm requires three inputs, the abdominal T2 and T1-FS images, provide a minimal set of five anatomical reference points (marker points) along the colonicand the colonic segmentation in T2 modality explained in Section 2.1. From these inputs, the algorithm trajectory to guide the segmentation algorithm. These points determined the start and end computes a non-rigid transformation (RegisterTrans fT2 →T1−FS ) to obtain the spatial corre- of each of the colon segments: ascending, transverse, descending, and pelvic. Depending spondence between T2 and T1-FS images. Next, an iterative registration process accurately on the complexity of the colonic shape, more points may be required to ensure a correct adapts the initial registered colon to T1-FS images. This iterative registration phase is segmentation result. From this information, the algorithm automatically computed the driven by a mesh deformation process based on a T1-FS content probabilistic model and colonic segmentation. The algorithm is founded on the estimation of the colon medial path mechanisms for preserving the T2 colon shape. and the usage of a probabilistic model for tissue classification based on the MRI intensities andSynchronized 2.3. the colon medial T2 and path distance. T1-FS A probability model guides a progressive segmenta- Inspection tion process based on a multigrid architecture using a 3D graph-cuts algorithm. This mul- As mentioned in Section 1, there may be confusing areas of the colon that require tigrid architecture provides computational scalability to the solution. the simultaneous visualization of both modalities to understand the colonic anatomy. Therefore, once both segmentations have been computed, their simultaneous visualization 2.2. Colon Segmentation in T1-FS Images and inspection allows the specialist to analyze, validate, and, if it is needed, to correct In Ref. [15] we presented an automatic colon segmentation algorithm for T1-FS im- the results. ages.ToThe algorithm achieve this,requires we usedthreethe inputs, non-rigidtheregistration abdominal T2 and T1-FS images, transformation and the computed in colonic 2.2 Section segmentation to obtain thein T2 modality spatial explained inbetween correspondence Section 2.1. T2 From these inputs, and T1-FS images.the algo- Specif- rithm given ically, computes a non-rigid a point transformation P in a T2 image, (RegisterTrans we can compute fT2→T1−FS) to point its correspondence obtaininthethespatial T1-FS correspondence space as: PT1−FS between T2 andfT2 = RegisterTrans T1-FS images. →T1−FS (P). Next, an Similarly, iterative from a registration point P in a process T1-FS ac- image, curately adapts the initial registered colon to T1-FS images. its correspondence point in the T2 space is: PT2 = RegisterTrans fT1→FS-T2 (P).This iterative registration phase is driven Figure by a snapshot 3 shows mesh deformation processinterface of the application based on inawhich T1-FStwocontent T2 and probabilistic T1-FS syn- model andimages chronized mechanisms for preserving of a segmented thevisualized colon are T2 colon shape. in two different windows. Every time the specialist clicks on a pixel of one image (T2 or T1-FS) the system automatically computes 2.3. Synchronized T2 and T1-FS Inspection tion 2.2 to obtain the spatial correspondence between T2 and T1-FS images. Specifically, given a point P in a T2 image, we can compute its correspondence point in the T1-FS space as: PT1−FS = RegisterTrans fT2→T1−FS(P). Similarly, from a point P in a T1-FS image, its corre- spondence point in the T2 space is: PT2 = RegisterTrans fT1→FS-T2(P). Figure 3 shows a snapshot of the application interface in which two T2 and T1-FS Diagnostics 2023, 13, 910 5 of 10 synchronized images of a segmented colon are visualized in two different windows. Every time the specialist clicks on a pixel of one image (T2 or T1-FS) the system automat- ically computes its corresponding pixel position in the other image (T1-FS or T2) by trans- its corresponding forming the pixelpixel position position usinginthe theformula other image (T1-FS or previously T2) by transforming presented. the pixel Moreover, using the position using the mouse wheel, theformula user canpreviously navigate presented. forward and Moreover, backward using the mouse around wheel,in the images the a user syn- can navigate chronous forward fashion and backward following the samearound scheme:thewhen imagestheinspecialist a synchronous movesfashion following the mouse wheel the same scheme: when the specialist moves the mouse wheel on one on one window, both views are updated accordingly to display the same location. Thewindow, both views are updated window accordingly directly affectedto by display the same the wheel location. is moved The window directly proportionally affected to the wheel by the displace- wheel ment, whereas the movement of the other window has to be calculated using the of is moved proportionally to the wheel displacement, whereas the movement the trans- other window haspresented. form previously to be calculated using the transform previously presented. Figure 3. Figure 3. Snapshot Snapshot of of the the synchronized synchronized T2 T2 and and T1-FS T1-FS module. module. On On the the left left side, side, an an original original T2 T2 image image is shown is shown with with the the colon colon segmentation segmentationoutlined outlinedusing usingaadifferent differentcolor colorfor foreach eachsegment. segment.On Onthe theright right side, the same visualization scheme is used with its corresponding original T1-FS side, the same visualization scheme is used with its corresponding original T1-FS image. image. The main The main novelty novelty of of this this inspection inspection functionality functionality is that,that, to to our our knowledge, knowledge, itit is is the the first one first one able able toto allow allow synchronous synchronous navigation navigation of of both both MRI MRI sequences sequences considering considering thethe deformation suffered in deformation in the the acquisition acquisition process. process. Note that the original original MRI MRI images images areare visualized without visualized without altering altering them, them, but but the interactive navigation allows computing the cor- responding responding image image using using thethe non-rigid non-rigid transformation transformation matrix. Moreover, the synchronized matrix. Moreover, synchronized inspection facilitates a 3D mental reconstruction of the colon in both inspection facilitates a 3D mental reconstruction of the colon in both MRI modalities,MRI modalities, which is helpful which to validate is helpful the segmentation to validate result inresult the segmentation both in sequences, and theand both sequences, medical expert the medical can modify expert the colon can modify thesegmentation, if needed, colon segmentation, either ineither if needed, T2 orin T1-FS T2 orindependently. The 3D T1-FS independently. simultaneous visualization The 3D simultaneous of both segmentations visualization of both segmentationsis also possible. However, is also possible. performing However, per- corrections in the 3Din forming corrections space is quite the 3D spacecomplex and medical is quite complex andspecialists prefer theprefer medical specialists 2D synchro- the 2D nized visualization synchronized of T2 andofT1-FS visualization images T2 and T1-FSsince imagesit provides since it detailed providesinformation about the detailed information anatomical structures around the colon and facilitates the validation of the boundaries in the automatically generated segmentations. 2.4. Colon Content Analysis Once the segmentation of the colon has been obtained in both sequences and has been validated by the specialist, the computation of the colonic content can be performed. This process consists of the classification of pixel intensities according to the percentage of solid content they represent. The solid percentage computation is based on a k-means clustering approach (details in [12]). Apart from fecal and gas content, other parameters related to the shape of the colon (length, perimeter, and radius) could also be computed from the colonic segmentation in T2 images (details in [16]). Diagnostics 2023, 13, 910 6 of 10 2.5. Acquisition Protocol The proposed framework has been used in independent clinical research studies. In all the experiments, each time a subject was scanned, two coronal image series of the abdomen were obtained: a T2-weighted HASTE (T2) sequence during two apneas of approximately 20 s each and a T1-weighted VIBE Fat-Sat (T1-FS) sequence in one apnea of 12 s. The imaging examinations were performed using two different 1.5-T MR imaging systems (Aera; Siemens Healthcare, Erlangen, Germany and Ingenia; Philips Medical System, Andover, USA) with abdominal coils wrapped around the abdomen. The subjects were placed in prone position. The specific acquisition protocols for the different studies were also different for each MR system (see [4,5,13] for a detail description). Images were obtained without oral or intravenous contrast, colonic infusion of gas or liquids or administration of antiperistaltic drugs. All images were codified and analyzed blind concerning the source, acquisition, and any preceding intervention by two trained physicians in MRI images under the supervision of an expert radiologist. 3. Results In this section, the experiments carried out to evaluate each module of the overall pipeline are presented. The experiments tested the accuracy obtained by the segmenta- tion algorithms, the computational and the user interaction cost, and the usability of all the modules. 3.1. Validation of the Colon Segmentation from T2 and T1-FS Images The validation of the colon segmentation from T2 images was presented in [7]. In total, 52 scans were used in the accuracy experiment: 30 scans from 15 healthy participants (8 women, 7 men) and 22 scans from 11 patients (8 women, 3 men) with irritable bowel syndrome (IBS); participants had a normal body mass index or were overweight (body mass index between 18.5 and 29.9 Kg/m2 ), i.e., no underweight or obese subjects were included. The assessment of the algorithm accuracy was performed by comparing the segmentation results to the ground truth (specialist manual segmentation) using the dice similarity coefficient (DSC). The achieved accuracy was (µDSC = 0.85, σDSC = 0.06), demonstrating its suitability for clinical usage. Figure 4 compares results of the colon segmentation computed by the algorithm with respect to a ground truth segmentation manually performed by the medical expert. In the left image, the medical expert did not realize that a small ileum region had been wrongly identified as colon by the segmentation algorithm. This misclassification was probably due to a similar pixel intensity in both the ileum and the adjacent colon, which made it very difficult to identify the colon in that area. As explained in Section 3.2, a simultaneous visualization of the T1-FS images could have helped him to fix this wrong identification. From the point of view of performance, the average execution times (in minutes) for all the volumes involved in the experiment were in the range 5.7 ± 1.16 min, with a maximum peak of 8.5 min, and the memory usage was below 2.5 GB. Focusing on the usability of the process, the experiments demonstrated that this approach could be considered as low demanding in terms of user time (on average 50 of usage time). Compared to the initial approach presented in [12], usage times have been reduced by 80%. Regarding the validation of the colon segmentation in T1-FS images, the algorithm accuracy was evaluated by comparing the automatic segmentation results to manual seg- mentations of the colonic fecal content (taken as ground truth) performed by the specialist (in [15] the validation experiments were presented). Note that the specialist only seg- mented the fecal content because the complete segmentation was too demanding and almost impossible in areas where no fecal content was presented (areas with gas). GB. Focusing on the usability of the process, the experiments demonstrated that this ap- proach could be considered as low demanding in terms of user time (on average 5′ of Diagnostics 2023, 13, 910 7 of 10 usage time). Compared to the initial approach presented in [12], usage times have been reduced by 80%. Figure 4. Comparison of segmentation results (in blue) with a ground truth manual colon segmenta- tion (in green) in T2 images. In the left image, the red arrow marks a small ileum region wrongly assigned to colon by the automatic algorithm. In the right image, a highly challenging area even for the specialists (outlined in red) is shown. Courtesy of [7]. Regarding the validation of the colon segmentation in T1-FS images, the algorithm accuracy was evaluated by comparing the automatic segmentation results to manual seg- Figure Figure 4. Comparison 4. Comparison of segmentation of segmentation results results (in(in blue) blue) withwith a ground a ground truth truth manual manual colon colon segmenta- segmenta- mentations of the colonic fecal content (taken as ground truth) performed by the specialist tion tion (in(in green) green) in in T2T2 images. images. In In thethe leftleft image, image, thetheredred arrow arrow marks marks a small a small ileum ileum region region wrongly wrongly (in [15] the assigned validation to colon by by thetheexperiments automatic were presented). algorithm. In Inthethe right Noteathat image, thechallenging specialistarea only seg- assigned to colon automatic algorithm. right image,highly a highly challenging area even forfor even mented thethe the specialists fecal content (outlined in because red) is the shown. complete Courtesy segmentation of [7]. was too demanding and al- specialists (outlined in red) is shown. Courtesy of [7]. most impossible in areas where no fecal content was presented (areas with gas). InRegarding total, 90 scans In total, the 90 obtained validation scans from obtained offrom35 colon the healthy volunteers segmentation 35 healthy volunteers (8inwomen, (8T1-FS women, 27 27 men; images, 20–28 the men; Kg/m algorithm 20–28 Kg/m2 2 bodybodymass accuracy index mass was range) index evaluated were range) used.used. were by comparing The the quantitative The quantitative automatic validation metric, validation segmentation named metric, results R, was R, to named manual de-was seg- fined as theas defined mentations percentage ofthe colonicof ground thepercentage fecal truthtruth of ground content fecal (taken content fecal inside content as ground the colon inside truth) segmentation the colon performed segmentationresult. result. by the specialist The achieved (inThe [15]achieved accuracy accuracy the validation was (µR(µ was = R96.2%, experiments = 96.2%, were σR presented). =σR4.2), demonstrating = 4.2), Note that its demonstrating theviability its foronly viability specialist clinical for clinical seg- usage (see usage the mented Figure (seefecal 5). Figure The mean 5). The content execution mean execution because time the complete was 9 min timesegmentation without was 9 min without any was too human intervention. anydemanding human intervention. and al- most impossible in areas where no fecal content was presented (areas with gas). In total, 90 scans obtained from 35 healthy volunteers (8 women, 27 men; 20–28 Kg/m2 body mass index range) were used. The quantitative validation metric, named R, was de- fined as the percentage of ground truth fecal content inside the colon segmentation result. The achieved accuracy was (µR = 96.2%, σR = 4.2), demonstrating its viability for clinical usage (see Figure 5). The mean execution time was 9 min without any human intervention. Figure 5. Comparison of segmentation results in a T1-FS coronal slice. In the left image, the original slice visualized with the corresponding automatic colon segmentation outlined using a different color for each of its segments. In the right image, the same coronal slice is visualized with the corresponding manual identification of the fecal content, used as a ground truth in the validation process. 3.2. Validation of the Synchronized Navigation in T2 and T1-FS Images As shown in previous sections, a synchronized visualization of T1-FS and T2 sequences is required to validate the coherence between both colon segmentations and, also, to correct the results in some confusing zones (see an example in Figure 6). responding manual identification of the fecal content, used as a ground truth in the validation pro- cess. 3.2. Validation of the Synchronized Navigation in T2 and T1-FS Images As shown in previous sections, a synchronized visualization of T1-FS and T2 se- Diagnostics 2023, 13, 910 8 of 10 quences is required to validate the coherence between both colon segmentations and, also, to correct the results in some confusing zones (see an example in Figure 6). Figure 6. Comparative Figure 6. Comparative images images ofof aa confusing confusing area area of of the the cecum cecum inin T2 T2 (left (left image) image) and and T1-FS T1-FS (right (right image) sequences. Note that in the T2 sequence, part of cecum was not included in the initial segmenta- image) sequences. Note that in the T2 sequence, part of cecum was not included in the initial tion because it was confused with the small bowel; a zoom (central image) provides a better visuali- segmentation because it was confused with the small bowel; a zoom (central image) provides a better zation of the non-included area (red line). However, in the synchronized visualization with the T1 visualization of image), sequence (right the non-included area (redthe the feces facilitate line). However, cecum in the synchronized demarcation. visualization Therefore, using with the new inspec- tion module, the medical specialist should be able to correct and include the cecum in the finalnew the T1 sequence (right image), the feces facilitate the cecum demarcation. Therefore, using the seg- inspection module, the medical specialist should be able to correct and include the cecum in the mentation. final segmentation. The validation of this module consisted of the visual analysis of some colon segmen- tationThe validation results of thisexperts. by medical module Weconsisted of the specially visual chose thoseanalysis cases of somethe where colon segmenta- isolated vali- tion results by medical experts. We specially chose those cases dation of T2 and T1-FS segmentation showed some small differences. where the isolated validation of T2 and T1-FS segmentation showed some small differences. In all the analyzed cases, when the specialist detected a complex area where the colon In all the analyzed cases, when the specialist detected a complex area where the colon was difficult to identify using only one sequence (T2 or T1-FS), the analysis of the same was difficult to identify using only one sequence (T2 or T1-FS), the analysis of the same area in the other sequence helped him to decide. In general, these confusing areas were area in the other sequence helped him to decide. In general, these confusing areas were located in the T2 images, especially where the small intestine was very close to the colon located in the T2 images, especially where the small intestine was very close to the colon and their pixel intensities were similar (see Figure 6). The new module facilitates and gives and their pixel intensities were similar (see Figure 6). The new module facilitates and gives security to the specialists at the time of determining the belonging of the area to the colon. security to the specialists at the time of determining the belonging of the area to the colon. Another complex area was the liver zone, where sometimes the separation between the Another complex area was the liver zone, where sometimes the separation between the colon and the liver is not clear enough to determine, accurately, the separation between colon and the liver is not clear enough to determine, accurately, the separation between both anatomical structures. The use of the synchronized navigation helped the specialist both anatomical structures. The use of the synchronized navigation helped the specialist to to be confident in this decision. be confident in this decision. After experimentingwith After experimenting withthis this new new module, module, somesome future future functionalities functionalities were were dis- discussed cussed in orderintoorder to improve improve their use. their use.experts Medical Medical experts agree that aagree usefulthat a useful functionality functionality could be the could synchronized correction in both sequences. In this way, the interactivethe be the synchronized correction in both sequences. In this way, interactive corrections cor- should rections should be only performed in one sequence to fix both colon be only performed in one sequence to fix both colon segmentations simultaneously. segmentations sim- ultaneously. 4. Discussion The computerized analysis of MRI images using the proposed algorithm has provided objective information on the volume and distribution of colonic content in normal condi- tions and in relation to abdominal symptoms [5]. Previous validation studies compared images before and after colonic infusion of known volumes of air, which proved the relia- bility and precision of the algorithm [13]. Further refinements and automatization of the algorithm have permitted large clinical studies comparing the effect of different dietary interventions on colonic content [4]. Other studies have applied the algorithm to quantify gas and non-gaseous colonic content in different digestive disorders and obtained unique information on the relation between colonic volumes and abdominal symptoms [17]. Diagnostics 2023, 13, 910 9 of 10 We wish to acknowledge several limitations of this study. First, we included in the study participants with normal body mass index or overweight, and the performance of the system in patients outside this range remains to be established; particularly, the accuracy may be reduced in underweight participants with a thin abdominal fat layer. Furthermore, the technique has been applied for the evaluation of different physiological conditions in healthy subjects and in patients with functional digestive disorders, but the applicability in case of organic disorders, such as inflammatory bowel disease, intestinal neuropathy, and postsurgical conditions, has not been validated; for instance, colonic morphology may be burred in case of excessive contractile activity. Our study may have clinical relevance. Indeed, the quasi-automatic current version of the algorithm has hugely accelerated the segmentation process and reduced the spe- cialist’s effort, while maintaining high accuracy and repeatability. Furthermore, the novel synchronized navigation module facilitates segmentation in those areas where it is difficult to distinguish the colon from other adjacent anatomical structures. Altogether, the improved platform, with an interactive tool for validation, could be suitable for the measurement of colonic volumes in daily clinical practice in the near future. Author Contributions: Conceptualization: B.O., E.M., I.N., Á.B. and F.A.; Software development: B.O. and E.M.; Á.B. analyzed the MRI images; data analysis: C.M. and F.A.; writing—original draft preparation: E.M. and I.N.; writing—review and editing: I.N., C.M. and F.A.; supervision: I.N., C.M. and F.A. All authors have read and agreed to the published version of the manuscript. Funding: This work was supported by the Spanish Ministry of Science and Innovation (Proyectos de Generación de Conocimiento), PID2021-122295OB-I00, and Agencia Estatal de Investigación and Fondos FEDER, PID2021-122136OB-C21); Ciberehd is funded by the Instituto de Salud Carlos III. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Not applicable. Acknowledgments: The authors thank Gloria Santaliestra for secretarial assistance and Pere-Pau Vazquez for his invaluable help in usable interface design. Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results. Abbreviations The following abbreviations are used in this manuscript: MRI Magnetic resonance imaging T2 T2-weighted HASTE sequence T1-FS T1-weighted VIBE fat-sat sequence References and Note 1. Accarino, A.; Perez, F.; Azpiroz, F.; Quiroga, S.; Malagelada, J. Intestinal Gas and Bloating: Effect of Prokinetic Stimulation. Gastroenterology 2008, 103, 2036–2042. [CrossRef] 2. Accarino, A.; Perez, F.; Azpiroz, F.; Quiroga, S.; Malagelada, J. Abdominal Distention Results From Caudo-ventral Redistribution of Contents. Gastroenterology 2009, 136, 1544–1551. [CrossRef] 3. Barba, E.; Quiroga, S.; Accarino, A.; Monclús, E.; Malagelada, C.; Burri, E.; Navazo, I.; Malagelada, J.; Azpiroz, F. Mechanisms of abdominal distension in severe intestinal dysmotility: Abdomino-thoracic response to gut retention. Neurogastroenterol. Motil. 2013, 25, e389–e394. [CrossRef] [PubMed] 4. Barber, C.; Mego, M.; Sabater, C.; Vallejo, F.; Bendezu, R.A.; Masihy, M.; Guarner, F.; Espín, J.C.; Margolles, A.; Azpiroz, F. 307 Differential Effects of Western and Mediterranean-Type Diets on Gut Microbiota: A Metagenomics and Metabolomics Approach. Nutrients 2021, 13, 2638. [CrossRef] [PubMed] 5. Bendezú, R.; Mego, M.; Monclus, E.; Merino, X.; Accarino, A.; Malagelada, J.; Navazo, I.; Azpiroz, F. Colonic content: Effect of diet, meals, and defecation. Neurogastroenterol. Motil. 2017, 29, e12930. [CrossRef] Diagnostics 2023, 13, 910 10 of 10 6. Clarke, L.; Velthuizen, R.; Camacho, M.; Heine, J.; Vaidyanathan, M.; Hall, L.; Thatcher, R.; Silbiger, M. MRI segmentation: Methods and applications. Magn. Reson. Imaging 1995, 13, 343–368. [CrossRef] 7. Orellana, B.; Monclús, E.; Brunet, P.; Navazo, I.; Bendezú, Á.; Azpiroz, F. A scalable approach to T2-MRI colon segmentation. Med. Image Anal. 2020, 63, 101697. [CrossRef] 8. Brown, R.W.; Cheng, Y.C.N.; Haacke, E.M.; Thompson, M.R.; Venkatesan, R. Magnetic Resonance Imaging: Physical Principles and Sequence Design, 2nd ed.; John Wiley&Sons: Hoboken, NJ, USA, 2014. [CrossRef] 9. Pritchard, S.E.; Marciani, L.; Garsed, K.; Hoad, C.; Thongborisute, W.; Roberts, E.; Gowland, P.A.; Spiller, R.C. Fasting and postprandial volumes of the undisturbed colon: Normal values and changes in diarrhea-predominant irritable bowel syndrome measured using serial MRI. Neurogastroenterol. Motil. 2014, 26, 124–130. [CrossRef] [PubMed] 10. Nilsson, M.; Sandberg, T.H.; Poulsen, J.L.; Gram, M.; Frøkjær, J.B.; Østergaard, L.R.; Krogh, K.; Brock, C.; Drewes, A.M. Quantification and variability in colonic volume with a novel magnetic resonance imaging method. Neurogastroenterol. Motil. 2015, 27, 1755–1763. [CrossRef] 11. Sandberg, T.H.; Nilsson, M.; Poulsen, J.L.; Gram, M.; Frøkjær, J.B.; Østergaard, L.R.; Drewes, A.M. A novel semi-automatic segmentation method for volumetric assessment of the colon based on magnetic resonance imaging. Abdom. Imaging 2015, 40, 2232–2241. [CrossRef] 12. Ceballos Inza, V.; Monclús Lahoya, E.; Vázquez Alcocer, P.P.; Bendezú García, Á.; Mego, M.; Merino, X.; Azpiroz Vidaur, F.; Navazo Álvaro, I. Semi-automatic colonic content analysis for diagnostic. In Proceedings of the EuroVis 2017: Eurographics/IEEE VGTC Conference on Visualization 2017, Posters, Barcelona, Spain, 12–16 June 2017; The Eurographics Association: Lower Saxony, Germany, 2017; pp. 121–123. [CrossRef] 13. Bendezú, A.; Barba, E.; Burri, E.; Cisternas, D.; Accarino, A.; Quiroga, S.; Monclús, E.; Navazo, I.; Malagelada, J.R.; Azpiroz, F. Colonic content in health and its relation to functional gut symptoms. Neurogastroenterol. Motil. 2016, 28, 849–854. [CrossRef] [PubMed] 14. Orellana, B.; Monclús, E.; Brunet, P.; Navazo, I.; Bendezú, A.; Azpiroz, F. Quasi-automatic Colon Segmentation on T2-MRI Images with Low User Effort. In Proceedings of the Medical Image Computing and Computer Assisted Intervention—MICCAI 2018 21st International Conference, Granada, Spain, 16–20 September 2018; pp. 638–647. 15. Orellana, B.; Monclús, E.; Brunet, P.; Navazo, I.; Bendezú, Á.; Azpiroz, F. Under review 2023. 16. Males, J.; Monclús, E.; Díaz, J.; Navazo, I.; Vázquez, P.P. Interactive framework for the visual exploration of colonic data. Comput. Graph. 2020, 91, 39–51. [CrossRef] 17. Malagelada, C.; Bendezú, R.A.; Seguí, S.; Vitrià, J.; Merino, X.; Nieto, A.; Sihuay, D.; Accarino, A.; Molero, X.; Azpiroz, F. Motor dysfunction of the gut in cystic fibrosis. Neurogastroenterol. Motil. 2020, 32, e13883. [CrossRef] Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.