Journal of Neurology https://doi.org/10.1007/s00415-022-11494-0 ORIGINAL COMMUNICATION Neural correlates of digital measures shown by structural MRI: a post‑hoc analysis of a smartphone‑based remote assessment feasibility study in multiple sclerosis Marco Ganzetti1 · Jennifer S. Graves2 · Sven P. Holm1 · Frank Dondelinger1,6 · Luciana Midaglia3,4 · Laura Gaetano1 · Licinio Craveiro1 · Florian Lipsmeier1 · Corrado Bernasconi1 · Xavier Montalban3,4 · Stephen L. Hauser5 · Michael Lindemann1 Received: 20 June 2022 / Revised: 15 November 2022 / Accepted: 17 November 2022 © The Author(s) 2022 Abstract Background A study was undertaken to evaluate remote monitoring via smartphone sensor-based tests in people with multi- ple sclerosis (PwMS). This analysis aimed to explore regional neural correlates of digital measures derived from these tests. Methods In a 24-week, non-randomized, interventional, feasibility study (NCT02952911), sensor-based tests on the Floodlight Proof-of-Concept app were used to assess cognition (smartphone-based electronic Symbol Digit Modalities Test), upper extremity function (Draw a Shape Test, Pinching Test), and gait and balance (Static Balance Test, Two- Minute Walk Test, U-Turn Test). In this post-hoc analysis, digital measures and standard clinical measures (e.g., Nine- Hole Peg Test [9HPT]) were correlated against regional structural magnetic resonance imaging outcomes. Seventy-six PwMS aged 18–55 years with an Expanded Disability Status Scale score of 0.0–5.5 were enrolled from two different sites (USA and Spain). Sixty-two PwMS were included in this analysis. Results Worse performance on digital and clinical measures was associated with smaller regional brain volumes and larger ventricular volumes. Whereas digital and clinical measures had many neural correlates in common (e.g., putamen, globus pallidus, caudate nucleus, lateral occipital cortex), some were observed only for digital measures. For example, Draw a Shape Test and Pinching Test measures, but not 9HPT score, correlated with volume of the hippocampus (r = 0.37 [drawing accu- racy over time on the Draw a Shape Test]/ − 0.45 [touching asynchrony on the Pinching Test]), thalamus (r = 0.38/ − 0.41), and pons (r = 0.35/ − 0.35). Conclusions Multiple neural correlates were identified for the digital measures in a cohort of people with early MS. Digital measures showed associations with brain regions that clinical measures were unable to demonstrate, thus providing potential novel information on functional ability compared with standard clinical assessments. Keywords Multiple sclerosis · Smartphone sensor · MRI · Neural correlates · Digital health technology tools 4 * Florian Lipsmeier Department of Medicine, Autonomous University florian.lipsmeier@roche.com of Barcelona, Barcelona, Spain 5 1 Department of Neurology, University of California San F. Hoffmann-La Roche Ltd, Basel, Switzerland Francisco, San Francisco, CA, USA 2 Department of Neurosciences, University of California San 6 Present Address: Novartis Institutes for Biomedical Research, Diego, San Diego, CA, USA Basel, Switzerland 3 Department of Neurology‑Neuroimmunology, Centre d’Esclerosi Múltiple de Catalunya (Cemcat), Hospital Universitari Vall d’Hebron, Barcelona, Spain 13 Vol.:(0123456789) Journal of Neurology Introduction San Francisco, CA) and Spain (Multiple Sclerosis Centre of Catalonia, Vall d’Hebron University Hospital, Barce- Multiple sclerosis (MS) is a chronic disease of the central lona); MRI data were only collected from PwMS. PwMS nervous system [1]. A long-held view of MS is of a multi- were diagnosed according to the 2010 revised McDonald focal, immune-mediated, inflammatory-demyelinating, and criteria [15] and had an Expanded Disability Status Scale degenerative disorder of the white matter. Recent struc- (EDSS) [16] at baseline of 0.0–5.5. A sample size of least tural magnetic resonance imaging (MRI) studies, however, 70 PwMS was considered adequate to detect a linear cor- have revived the important role of cortical pathology in relation coefficient of 0.33 with > 80% power. the pathophysiology of MS and in the rate of disability progression [2]. Floodlight PoC app MRI-measured atrophy occurs in distinct, non-random patterns and involves many different regions, including At baseline, all participants received a pre-configured smart- the cerebral cortex, deep gray matter, brainstem, cerebel- phone (Samsung Galaxy S7) with the Floodlight PoC app lum, and spinal cord [2–7]. The rate of atrophy differs from pre-installed [13, 14]. The app prompted all participants to region to region [6]. However, the progression of regional perform daily and weekly sensor-based tests of functional atrophy tends to follow specific sequences depending on ability (referred to as “active tests”) at home without super- the MS phenotype. In relapse-onset MS, the first regions vision by a physician or nurse, in three key domains affected affected by atrophy include the posterior cingulate cortex by MS. Cognition was assessed by the smartphone-based and precuneus, while the globus pallidus and medial precen- electronic Symbol Digit Modalities Test (e-SDMT), upper tral gyrus are among the last regions to become atrophic. In extremity function by the Draw a Shape Test and Pinching primary progressive MS, by comparison, the first regions to Test, and gait and balance by the Static Balance Test (SBT), become atrophic are the thalamus, cuneus, and precuneus, Two-Minute Walk Test (2MWT), and U-Turn Test (UTT) and the last regions to show signs of atrophy are the frontal (Fig. 1). Performance on each active test was quantified by operculum and middle temporal gyrus [7]. a set of digital measures extracted from the sensor signals Although there have been efforts to link clinical dis- that are illustrative of the test (Table 1). ability to regional atrophy [8–12], the relationship between digital measures of functional ability and MRI outcomes Smartphone sensor‑based tests has not been extensively studied. Previously we showed that digital measures obtained with the Floodlight Proof- Key neurologic functions underlying cognitive information of-Concept (PoC) app in people with MS (PwMS) cor- processing speed were assessed by the e-SDMT. It instructed relate with global MRI outcomes, in particular with total participants to correctly match as many symbols as possible brain volume [13]. However, this analysis did not con- to their paired digits within 90 s, according to a symbol–digit sider the regional specificity of cortical, deep gray matter, key also displayed on the smartphone screen. infratentorial, and spinal cord pathology. Here we extend The Draw a Shape Test assessed fine finger movements, this prior work by assessing the regional neural correlates or dexterity; participants were prompted to draw six pre- of an expanded set of digital measures. written shapes of increasing complexity (two diagonal lines, a square, a circle, a figure-of-8, and a spiral). By compari- son, the Pinching Test measured fine pinching or grasp- ing dexterity. Participants were instructed to successfully Methods pinch as many circular tomatoes appearing on the screen as possible within 30 s, with new tomatoes appearing at Study design and participants random locations. Finally, gait and balance were assessed by three active This 24-week, non-randomized, interventional, feasibil- tests: the SBT, 2MWT, and UTT. The SBT required par- ity study (clinicaltrials.gov: NCT02952911) aimed to ticipants to stand as still as possible for 30 s. The 2MWT assess the feasibility of remotely monitoring PwMS with assessed regular, straight walking for 2 min. The UTT, which the Floodlight PoC app, through sensor-based assessment measured both dynamic balance and gait, instructed partici- via a provisioned smartphone device [14]. The study pants to perform five consecutive U-turns that were at least design, inclusion, and exclusion criteria have been previ- 4 m apart, within a minute. The use of an assistive device ously reported [13, 14]. The study enrolled 76 PwMS and and/or orthotic was permitted as needed while performing 25 healthy controls aged 18–55 years, from two separate the 2MWT and UTT. sites in the USA (University of California San Francisco, 13 Journal of Neurology Fig. 1  Screenshots of the smartphone sensor-based tests included in a Shape Test, Pinching Test, Static Balance Test, Two-Minute Walk the Floodlight PoC app. Screenshots depict the instruction screens for Test, and U-Turn Test. e-SDMT smartphone-based electronic Symbol the smartphone-based electronic Symbol Digit Modalities Test, Draw Digit Modalities Test, PoC Proof-of-Concept Digital data processing FLAIR (TR = 5000 ms; TE = 389 ms; TI = 1,800 ms; voxel size = 1 × 1 × 1 mm) sequences acquired in a Siemens Skyra. Since the active tests were unsupervised, quality control flags as defined in Montalban et al. [13] were applied to identify tests that were not performed in accordance with Image processing the test’s instructions and to ensure accurate interpretation of the collected data. This included flags to identify attempts MR images were analyzed with icobrain ms v.5.0 (icometrix, of “play to quit” (i.e., trying to complete an active test as Leuven, Belgium), a CE- and Food and Drugs Adminis- quickly as possible without performing the instructed tasks) tration-certified software as medical device for automatic or instances where the smartphone device was kept on a labeling, visualization, and volumetric quantification of table during the gait and balance tests. Any test identified segmentable brain structures [17–20], which combines two by one of these flags was excluded from the analyses. All sequential pipelines. (1) An automated method for white remaining tests were considered as valid tests. Only partici- matter lesion segmentation that uses 3D T1-weighted and pants who contributed on average at least 1.5 valid tests per FLAIR MR images in a probabilistic model. The accuracy week (corresponding to 21% adherence) were included in and reproducibility of this software has been shown to be the analysis. comparable to other well-established MS lesion segmenta- tion algorithms (e.g., Lesion-TOADS, Lesion Segmentation Tool [17, 18]); (2) An automated multi-atlas cortical and MRI acquisition and processing subcortical segmentation method, which showed similar accuracy and reproducibility to commonly used automatic Image acquisition segmentation tools [21, 22]. Icobrain ms allowed extraction of normalized volumes for 34 anatomical regions, includ- Brain MRI scans were collected from PwMS at baseline ing cerebral gray matter regions (n = 20), deep gray mat- and at week 24 with Siemens 3-Tesla scanners (Siemens ter regions (n = 7), brainstem (n = 3), cerebellum (n = 2), Healthcare GmbH, Erlangen, Germany) following opti- lateral ventricles (n = 1), and cerebral white matter (n = 1). mized clinical practice protocols in place at the two inves- In addition, icobrain ms allowed calculation of the corpus tigating centers: (1) Hospital Universitari Vall d'Hebron, callosum area (n = 1) and upper cervical cord area (n = 1). three-dimensional (3D) T1-weighted (repetition time As the cervical spinal cord was within the field of view on [TR] = 2300 ms; echo time [TE] = 2.98 ms; inversion time brain scans, the mean upper cervical cord area along C2–C3 [TI] = 900 ms; voxel size = 1 × 1 × 1 mm) fluid-attenuated levels was computed from the brain images [23, 24] using inversion recovery (FLAIR; TR = 6000 ms; TE = 394 ms; the icobrain ms fully automated pipeline [25, 26] based on TI = 2100 ms; voxel size = 1 × 1 × 1 mm) sequences the Spinal Cord Toolbox [27] (see Table 1 for more details). acquired in a Siemens Trio Tim; (2) University of Cali- After visual inspection of the segmentation output, no par- fornia San Francisco, 3D T1-weighted (TR = 2300 ms; ticipants were excluded from the analyses due to low quality TE = 2.98 ms; TI = 900 ms; voxel size = 1 × 1 × 1 mm) MRI data or unsuccessful segmentation outputs. 13 Journal of Neurology Table 1  Outcome variables Floodlight PoC app Functional domain Smartphone Functional subdomain Digital ­measurea Higher scores sensor-based indicate performance active test that is: Upper extremity function Draw a Shape Test Accuracy Mean trace accuracy Better Swiftness of movement Mean trace celerity, 1/s Better Smoothness of movement CV of linear, angular, and radial Worse drawing velocity Pinching Test Accuracy Total number of successful pinches, n Better Responsiveness Gap time between pinch attempts, s Worse Pinching asynchrony Double touch asynchrony of successful Worse pinches, s Cognition e-SDMT Information processing speed Number of correct responses, n Better Maximum gap time between correct Worse responses, s Cognitive fatigue Speed fatigability index of the last 30 s Better Gait and balance SBT Postural control Sway path, m/s2 Worse UTT​ Turning ability (dynamic Turn speed, rad/s Better balance and gait) 2MWT Gait pace Step frequency, Hz Better Gait variability Step frequency variance, H­ z2 Worse 2 3 Gait intensity Step power, m­ /s Better ­ Lb Normalized volumes derived from brain MRI scans, m Cerebral cortex gray matter Deep gray matter Brainstem and upper cervical cord area Other 1. Middle frontal 1. Thalamus 1. Midbrain 1. Lateral ventricles 2. Superior frontal 2. Hippocampus 2. Pons 2. Cerebral white matter 3. Lateral and medial orbitofrontal 3. Putamen 3. Medulla oblongata 3. Corpus callosum area 4. Precentral 4. Globus pallidus 4. Upper cervical cord ­areac 4. Cerebellum—hemispheres 5. Postcentral 5. Caudate nucleus 5. Cerebellum—vermis 6. Paracentral 6. Amygdala 7. Opercularis, triangularis, orbitalis 7. Accumbens 8. Insula 9. Precuneus 10. Cuneus, pericalcarine, lingual 11. Anterior cingulate 12. Posterior cingulate 13. Isthmus cingulate 14. Parahippocampus, fusiform, entorhinal 15. Superior parietal 16. Supramarginal 17. Inferior parietal 18. Lateral occipital 19. Inferior and middle temporal 20. Superior and transverse temporal 2MWT Two-Minute Walk Test, e-SDMT smartphone-based electronic Symbol Digit Modalities Test, PoC Proof-of-Concept, SBT Static Balance Test, UTT​U-Turn Test a Computational definitions are provided in Table SI1 b Volumes were calculated for all regions with the exception of the corpus callosum and the upper cervical cord for which areas measured in ­mm2 were calculated c Upper cervical cord area was determined as mean area along C2–C3 levels 13 Journal of Neurology Standard clinical measures module for data analysis and scientific computing (www.​ stats​models.​org). During three clinical visits (baseline, week 12, and week 24) PwMS underwent clinical evaluation. These included the oral Symbol Digital Modalities Test (SDMT), Nine-Hole Results Peg Test (9HPT), Berg Balance Scale, Timed 25-Foot Walk (T25FW), and EDSS. The study enrolled 76 PwMS between November 28, 2016 and November 13, 2017 (study completion date was May Statistical analysis 4, 2018), of which 62 (82%) PwMS were included in the analyses presented here, comprising of 11 PwMS from Due to the exploratory nature of this analysis, an unbi- the University of California San Francisco and 51 PwMS ased approach with no pre-specified hypotheses was from the Hospital Universitari Vall d'Hebron. Of the 14 adopted to investigate associations between the brain/spi- excluded PwMS, 12 were excluded for poor adherence to nal cord regions and the digital measures obtained with the the active tests and two for missing MRI scans. Full base- Floodlight PoC app (Table 1) and the standard clinical meas- line demographics and disease characteristics have been ures. Two separate cross-sectional analyses were conducted previously published [14] and those included in the current on the digital measures and, for comparison, on the stand- analyses are provided in Table 2. Of those PwMS included, ard clinical measures. First, these measures were correlated 68% were female, mean age was 39.7 years (stand- against regional, structural MRI outcomes using Spearman’s ard deviation [SD] 7.5), mean EDSS at baseline was 2.5 rank correlation. Since the digital measures may be con- (SD 1.4; range 0.0–5.5), and 89% were diagnosed with founded by age, sex, and body mass index, the correlation relapsing–remitting MS. In order to exclude any potential analysis was adjusted for these confounders with a robust effect of the MS phenotype on volumetric MRI data, we linear model via iteratively reweighted least squares [28]. compared all 36 regions of interest between the groups of Statistical significance was set at q < 0.05 after applying the people with relapsing–remitting MS and progressive MS. As false discovery rate (FDR) correction to correct for multiple reported in Fig. SI2, significant differences were observed comparisons for each digital and standard clinical measure in only one of the 36 regions (combined occipital regions separately [29]. of cuneus, lingual gyrus, and pericalcarine cortex). This In the second analysis, the variance in the digital meas- suggests that regional volumes were not dissimilar across ures that can be explained by volumetric MRI data was esti- relapsing and progressive MS phenotypes in this cohort. mated. Here, a Bayesian ridge regression model [30] with Additionally, no significant changes between the two study leave-one-out cross validation was used to estimate the sites were found for age (p = 0.846), sex (p = 0.145), body R2 score, i.e., a measure related to the proportion of the vari- mass index (p = 0.905), and disease phenotype (p = 0.436) ance in the dependent variable (digital measure or standard (Fig. SI3). clinical measure) that is predictable from the independent The regional neural correlates of the digital measures and variables (structural MRI outcomes). The following two the standard clinical measures are summarized in Fig. 2. models were considered: (1) the univariate Whole Brain Uncorrected p values and the respective FDR-corrected model, which included the normalized total brain volume p values (q values) are reported in the supplementary to estimate the variance in the digital measures and standard appendix (Fig. SI4 and SI5). Brain maps highlighting the clinical measures; (2) the multivariate Parcellation model, statistically significant correlations after FDR correction which instead included the normalized volume and cross- are provided in Fig. 3. Higher EDSS scores correlated with sectional area measurements of 36 MRI regions. smaller volumes of the cerebral white matter (r =  − 0.42, Given the relatively short duration of the study (24 weeks) q < 0.05), lateral occipital lobe (r =  − 0.43, q < 0.05), insula and the stability of the clinical and MRI measures over the (r =  − 0.37, q < 0.05), putamen (r =  − 0.36, q < 0.05), and study period (Fig. SI1), a data aggregation approach was fol- globus pallidus (r =  − 0.38, q < 0.05). lowed to reduce variability and to deal more effectively with The e-SDMT and oral SDMT showed a similar corre- missing data. For both analyses, the data were aggregated lation pattern with regional MRI outcomes (Figs. 2, 3a). as follows: for the digital measures, the median across all On both, a lower number of correct responses was signifi- valid active tests was calculated; for the standard clinical cantly associated with smaller volumes of the cerebral white measures and MRI outcomes, the mean of the three clinical matter (r = 0.49 for e-SDMT/r = 0.47 for oral SDMT), cor- visits (baseline, week 12, and week 24) and two MRI scans pus callosum (r = 0.39/0.40, q < 0.05), postcentral gyrus (baseline and week 24), respectively, were calculated. (r = 0.33/0.34, q < 0.05), lateral occipital lobe (r = 0.38/0.30, All statistical analyses were performed in Python 3.7.9 q < 0.05), insula (r = 0.35/0.39, q < 0.05), superior and trans- (www.p​ ython.o​ rg) using statsmodels 0.12.0, an open-source verse temporal gyrus (r = 0.37/0.48, q < 0.05), paracentral 13 Journal of Neurology Table 2  Baseline demographics and disease characteristics pericalcarine cortex, lingual gyrus (r = 0.34, q < 0.05), and Variable PwMS cohort included between the e-SDMT maximum gap duration between cor- in the analyses (n = 62) rect responses and the volume of the superior frontal lobe (r =  − 0.37, q < 0.05). Site 11 (USA), 51 (Spain) Likewise, for each regional MRI outcome that showed Female, n/N (%) 42/62 (67.7) significant correlation with the 9HPT there was at least one Age, years, mean (SD) Draw a Shape Test measure that was significantly correlated Mean (SD) 39.7 (7.5) with the same MRI outcome (Figs. 2, 3b, SI6a). Worse per- Min.–Max. 20–57 formance on either test was associated with smaller volumes BMI, kg/m2 of the cerebral white matter (│r│ = 0.33–0.41, q < 0.05), Mean (SD) 24.4 (4.4) lateral occipital lobe (│r│ = 0.39–0.47, q < 0.05), and Min.–Max. 17.1–37.6 deep gray matter nuclei such as the putamen, globus pal- Diagnosis, n (%) lidus, caudate nucleus, and accumbens (│r│ = 0.31–0.44, RRMS 55 (89) q < 0.05), as well as a larger volume of the lateral ventricles PPMS 3 (5) (│r│ = 0.38–0.45, q < 0.05). However, some correlates were SPMS 4 (6) only observed on the Draw a Shape Test, but not the 9HPT. Disease duration at baseline, years For example, mean trace celerity correlated with volumes Mean (SD) 9.5 (6.6) of the thalamus (r = 0.38, q < 0.05), hippocampus (r = 0.37, Min.–Max. 0.7–24.9 q < 0.05), and isthmus cingulate (r =  − 0.37, q < 0.05). Normalized total brain volume, mL Velocity-based measures of round shapes correlated with Mean (SD) 1471.9 (65.0) volumes of the thalamus (│r│ = 0.35–0.43, q < 0.05), hip- Min.–Max. 1277.1–1628.1 pocampus (│r│ = 0.32–0.40, q < 0.05), and cuneus, perical- EDSS carine cortex, and lingual gyrus (│r│ = 0.38–0.47, q < 0.05). Mean (SD) 2.5 (1.4) A significant association was also observed between vari- Min.–Max. 0.0–5.5 ability of linear drawing velocity on the figure-of-8 and the Oral SDMT, number of correct responses cervical spinal cord area (r =  − 0.44, q < 0.05). Mean (SD) 53.4 (11.7) Similarly, for each correlation observed with the 9HPT, Min.–Max. 26–77 at least one Pinching Test measure was significantly associ- 9HPT, seconds ated with the same MRI outcome, except for the cerebral Mean (SD) 22.4 (4.3) white matter (Figs. 2, 3b). Some correlations, however, Min.–Max. 16.4–40.3 were observed only with the Pinching Test. This includes T25FW, seconds correlations between longer double touch asynchrony and Mean (SD) 6 (2) smaller volumes of the pons (r =  − 0.35, q < 0.05), thalamus Min.–Max. 3.6–12.5 (r =  − 0.41, q < 0.05), hippocampus (r =  − 0.45, q < 0.05), Berg Balance Scale, total score and interestingly with a larger volume of the anterior cin- Mean (SD) 52.4 (5.8) gulate cortex (r = 0.36, q < 0.05). Additionally, longer Min.–Max. 31–56 gap duration between pinches was also associated with a 9HPT Nine-Hole Peg Test, BMI body mass index, EDSS Expanded smaller volume of the thalamus (r =  − 0.36, q < 0.05), insula Disability Status Scale, PPMS primary progressive multiple scle- (r =  − 0.32, q < 0.05), and the cuneus, pericalcarine cortex, rosis, PwMS people with multiple sclerosis, RRMS relapsing–remit- and lingual gyrus (r =  − 0.35, q < 0.05). ting multiple sclerosis, SD standard deviation, SDMT Symbol Digit Modalities Test, SPMS secondary progressive multiple sclerosis, On the SBT, larger sway path correlated with a smaller T25FW Timed 25-Foot Walk volume of the paracentral lobule (r =  − 0.42, q < 0.05; Figs. 2, 3c). In contrast, a lower total score on the Berg Balance Scale (worse ability to balance) correlated mostly gyrus (r = 0.35/0.36, q < 0.05), and deep gray matter struc- with smaller volumes of deep gray matter structures tures such as the thalamus, putamen, globus pallidus, (r = 0.39 − 0.44 for correlations with q < 0.05; Figs. 2, 3c). and caudate nucleus (r = 0.37–0.51/0.51–0.60, q < 0.05). No correlations with any of the regional MRI outcomes A lower number of correct responses on either test was were found for any of the gait tests, neither for the digital also associated with a larger volume of the lateral ventri- measures derived from the 2MWT/UTT nor for the T25FW cles (r =  − 0.50/ − 0.57, q < 0.05). Two correlations were (Fig. 2; all q ≥ 0.05). only observed on the e-SDMT, but not on the oral SDMT. Next, the variance observed in the digital measures that This included the correlation between the e-SDMT num- can be explained by structural MRI outcomes was estimated ber of correct responses and the volume of the cuneus, using a Bayesian ridge regression model. Up to a third of the 13 Journal of Neurology Cerebral white matter 0.49 -0.42 0.30 0.38 0.23 -0.41 -0.33 -0.41 -0.30 -0.30 0.32 -0.05 -0.08 -0.09 -0.05 0.18 -0.40 0.47 0.33 -0.24 -0.42 Lateral ventricles -0.50 0.46 -0.43 -0.20 -0.40 0.45 0.40 0.38 0.46 0.46 -0.38 0.03 0.16 0.09 0.04 -0.19 0.39 -0.57 -0.25 0.30 0.29 Cerebellum – hemispheres 0.02 -0.08 0.08 0.04 0.22 -0.20 -0.26 -0.09 -0.19 0.01 0.10 0.14 0.04 -0.07 0.19 0.18 -0.09 0.10 0.20 -0.13 -0.12 Cerebellum – vermis -0.01 -0.08 0.09 0.19 0.20 -0.19 -0.25 -0.25 -0.24 -0.01 0.23 0.06 0.07 -0.09 0.20 0.06 -0.19 0.02 0.19 -0.11 -0.23 Midbrain 0.08 -0.13 0.03 -0.04 0.21 -0.13 -0.09 -0.03 -0.30 -0.14 0.14 0.19 -0.18 -0.19 0.16 -0.09 -0.13 0.10 0.09 0.10 -0.08 Pons 0.12 -0.04 0.06 0.02 0.35 -0.32 -0.31 -0.16 -0.35 -0.06 0.16 0.08 -0.12 -0.14 0.19 0.10 -0.23 0.27 0.12 -0.13 -0.10 Medulla oblongata -0.01 -0.12 -0.03 0.00 0.05 -0.15 -0.08 -0.02 -0.02 0.17 -0.06 0.17 0.08 -0.19 0.08 -0.05 -0.08 -0.13 0.09 -0.00 -0.04 Corpus callosum area 0.39 -0.28 0.22 0.27 0.21 -0.28 -0.21 -0.28 -0.21 -0.20 0.19 -0.16 -0.15 -0.04 -0.10 0.04 -0.24 0.40 0.07 -0.13 -0.24 Upper cervical cord area 0.25 -0.20 0.08 0.25 0.21 -0.44 -0.22 -0.29 -0.14 0.00 0.05 -0.15 0.20 -0.07 -0.15 0.11 -0.31 0.20 0.24 -0.17 -0.32 Opercularis, triangularis, orbitalis 0.05 -0.09 -0.00 -0.11 0.10 0.12 0.02 0.01 -0.16 -0.13 0.05 0.02 -0.04 -0.12 0.07 0.11 -0.04 0.07 0.04 -0.14 0.09 Middle frontal 0.16 -0.26 0.17 0.07 -0.13 -0.11 -0.13 -0.01 -0.06 -0.16 0.18 -0.21 -0.03 -0.13 0.05 -0.05 -0.11 0.15 0.09 -0.10 -0.05 Superior frontal 0.28 -0.37 0.34 0.16 -0.09 -0.14 -0.13 -0.12 -0.11 -0.22 0.25 -0.17 0.04 -0.11 0.06 0.12 -0.18 0.27 0.21 -0.18 -0.13 Precentral 0.13 -0.22 0.20 -0.01 -0.05 -0.01 -0.03 0.02 0.08 -0.07 0.04 -0.16 0.03 0.09 0.22 0.16 0.03 0.22 0.04 -0.10 0.04 Postcentral 0.33 -0.39 0.24 0.05 0.02 -0.21 -0.18 -0.09 -0.16 -0.31 0.18 -0.09 -0.11 -0.05 0.05 0.04 -0.18 0.34 0.16 -0.14 -0.13 Supramarginal 0.15 -0.26 0.20 -0.03 -0.02 -0.03 -0.05 0.08 -0.04 -0.07 0.14 -0.07 -0.05 -0.06 0.02 0.02 0.02 0.23 0.11 -0.07 -0.07 Superior parietal 0.12 -0.22 0.03 0.23 -0.24 -0.05 0.09 -0.04 -0.13 -0.10 0.13 0.15 -0.07 -0.07 0.24 -0.27 -0.11 0.12 -0.08 0.21 -0.07 Inferior parietal 0.12 -0.11 0.07 0.10 -0.07 0.00 0.07 -0.01 -0.03 0.01 0.06 0.05 -0.08 -0.14 0.14 -0.13 -0.02 0.17 -0.01 0.10 -0.08 Lateral occipital 0.38 -0.39 0.26 0.39 -0.00 -0.28 -0.10 -0.45 -0.32 -0.35 0.31 0.02 0.16 0.14 0.22 0.12 -0.47 0.30 0.16 -0.06 -0.43 Insula 0.35 -0.27 0.30 0.00 0.13 -0.21 -0.2 -0.16 -0.15 -0.32 0.19 -0.09 0.05 -0.27 -0.11 0.29 -0.25 0.39 0.37 -0.28 -0.37 Superior and transverse temporal 0.37 -0.33 0.27 -0.03 0.30 -0.24 -0.34 -0.08 -0.21 -0.25 0.31 -0.09 0.04 -0.21 -0.06 0.20 -0.14 0.48 0.21 -0.25 -0.19 Inferior and middle temporal 0.15 -0.18 0.17 -0.05 0.13 -0.03 -0.13 0.03 -0.13 -0.20 0.23 -0.08 0.01 -0.05 0.06 0.14 0.03 0.14 0.00 -0.14 0.12 Lateral and medial orbitofrontal -0.04 -0.04 -0.05 -0.09 -0.14 0.04 0.05 0.08 0.10 -0.07 0.10 -0.08 -0.01 -0.15 -0.12 -0.04 0.08 -0.07 0.04 -0.03 0.04 Anterior cingulate -0.28 0.27 -0.18 -0.22 -0.16 0.25 0.15 0.21 0.36 0.25 -0.24 -0.07 0.20 -0.00 -0.11 0.09 0.25 -0.36 0.07 -0.14 -0.00 Posterior cingulate -0.14 0.04 -0.18 -0.08 -0.18 0.17 0.13 0.15 0.19 0.16 -0.08 0.05 0.10 0.27 0.26 -0.04 0.06 -0.15 -0.04 0.00 0.14 Paracentral 0.35 -0.33 0.23 0.31 0.03 -0.39 -0.29 -0.37 -0.08 -0.29 0.21 -0.42 -0.11 0.13 -0.02 0.17 -0.22 0.36 0.24 -0.28 -0.21 Isthmus cingulate -0.22 0.17 -0.10 -0.04 -0.37 0.11 0.12 0.10 0.21 0.15 -0.33 -0.09 -0.04 0.17 -0.08 0.08 0.22 -0.26 -0.07 0.08 0.13 Precuneus 0.12 -0.11 0.13 0.11 -0.07 -0.29 -0.24 -0.20 -0.01 -0.09 -0.04 -0.24 -0.04 0.09 -0.11 0.13 0.02 0.10 0.23 -0.31 -0.07 Cuneus, pericalcarine, lingual 0.34 -0.28 0.27 0.32 0.15 -0.40 -0.38 -0.47 -0.27 -0.35 0.30 -0.13 0.05 0.13 -0.10 0.33 -0.26 0.27 0.27 -0.29 -0.20 Parahip., fusiform, entorhinal 0.03 -0.04 0.13 -0.14 0.14 -0.00 -0.13 0.12 -0.11 -0.11 0.07 0.00 -0.11 -0.25 -0.06 0.05 0.05 0.08 0.04 -0.07 0.09 Thalamus 0.51 -0.51 0.37 0.28 0.38 -0.43 -0.43 -0.35 -0.41 -0.36 0.32 -0.21 -0.01 0.00 -0.11 0.26 -0.30 0.60 0.42 -0.33 -0.32 Hippocampus 0.18 -0.22 0.06 0.18 0.37 -0.40 -0.40 -0.32 -0.45 -0.17 0.09 0.09 -0.04 -0.06 -0.09 0.13 -0.29 0.24 0.16 -0.17 -0.12 Putamen 0.44 -0.43 0.39 0.24 0.25 -0.43 -0.36 -0.38 -0.42 -0.42 0.30 -0.20 -0.10 -0.03 -0.15 0.18 -0.44 0.51 0.44 -0.30 -0.36 Globus pallidus 0.44 -0.40 0.33 0.30 0.21 -0.42 -0.36 -0.44 -0.44 -0.46 0.35 -0.23 -0.15 -0.11 -0.19 0.17 -0.43 0.52 0.39 -0.25 -0.38 Caudate nucleus 0.37 -0.35 0.28 0.18 0.23 -0.35 -0.32 -0.32 -0.43 -0.41 0.30 -0.08 0.06 -0.04 -0.12 0.15 -0.41 0.53 0.44 -0.32 -0.31 Amygdala 0.19 -0.21 0.30 -0.06 0.18 -0.15 -0.21 0.04 -0.10 -0.10 0.05 -0.21 0.02 -0.07 -0.00 0.17 -0.05 0.28 0.11 -0.23 -0.05 Accumbens 0.29 -0.23 0.22 0.12 0.25 -0.33 -0.31 -0.26 -0.46 -0.44 0.27 -0.02 -0.18 -0.13 -0.05 0.03 -0.39 0.44 0.31 -0.23 -0.22 Sway path EDSS SFI 30 Pinching gap time Step power mean Oral SDMT Berg Balance Scale 9HPT T25FW Circle CV angular vel. Turn speed mean Num. correct responses (CR) Max. gap time btw CR Mean trace accuracy Mean trace celerity Num. successful pinches Step frequency mean Figure-of-8 CV linear vel. Spiral CV radial vel. Step frequency variance Double touch asynchrony measures Draw a SBT UTT Shape Test Test 2MWT e-SDMT Clinical Pinching Spearman’s rho –1.00 –0.75 –0.50 –0.25 0.00 0.25 0.50 0.75 1.00 Fig. 2  Spearman’s rank correlation analysis of digital measures and EDSS, 9HPT, and T25FW, as well as for digital measures assessing standard clinical measures with global and regional MRI outcomes. max. gap time between correct responses on the e-SDMT; CV linear, Statistically significant (q < 0.05) positive and negative correlations angular, and radial velocity on the Draw a Shape Test; double touch are highlighted in red and blue, respectively. FDR correction was asynchrony and pinching gap time on the Pinching Test; sway path applied for each digital and standard clinical measure separately to on the SBT; and step frequency variance on the 2MWT. 2MWT Two- correct for multiple comparisons. Higher values equate to better per- Minute Walk Test, 9HPT Nine-Hole Peg Test, btw between, CR cor- formance on the oral SDMT and Berg Balance Scale, as well as for rect responses, CV coefficient of variation, EDSS Expanded Disabil- digital measures assessing number of correct responses and SFI 30 on ity Status Scale, e-SDMT smartphone-based electronic Symbol Digit the e-SDMT; trace accuracy and trace celerity on the Draw a Shape Modalities Test, FDR false discovery rate, num. number of, parahip. Test; number of pinches on the Pinching Test; mean step power and parahippocampus, SBT Static Balance Test, SDMT Symbol Digit mean step frequency on the 2MWT; and mean turn speed on the Modalities Test, SFI speed fatigability index, T25FW Timed 25-Foot UTT. In contrast, higher values equate to worse performance on the Walk, UTT​U-Turn Test, vel. drawing velocity 13 Journal of Neurology variance observed in the digital measures (R2 = 34%) could be explained when using either normalized total brain vol- ume as individual predictor (“Whole Brain” model) or volu- a metric data from the 36 individual MRI regions as multiple responses predictors (“Parcellation” model). In the upper extremity Correct function domain, however, comparable or higher R2 values were obtained with the Parcellation model compared with the Max. gap time e-SDMT btw correct Whole Brain model. This suggests that multiple predictors, responses thus multiple brain regions, can explain as much or more of the observed variance in the digital measures (Figs. 4, SI6b). The increase in R2 was most noticeable for mean trace celer- SFI 30 ity (R2 = 0.23 vs. 0), spiral coefficient of variation (CV) linear drawing velocity (R2 = 0.22 vs. 0.02), figure-of-8 CV linear Oral SDMT drawing velocity (R2 = 0.19 vs. 0.12), and double-touch asyn- responses Correct chrony (R2 = 0.16 vs. 0.06, respectively, for the Parcellation vs. Whole Brain model). This was also reflected by a less uniform distribution of the individual Bayesian ridge regression coef- b ficients across the 36 MRI regions (Fig. SI7). By compari- Mean trace accuracy son, the R2 score was more comparable across the two models for the standard clinical measure 9HPT (R2 = 0.28 vs. 0.25) (Fig. 4). Spiral CV Figure-of-8 CV Mean trace celerity Draw a Shape Test Discussion linear velocity Digital measures captured by the Floodlight PoC app demonstrated robust correlations with volumetric meas- velocity urements of specific brain regions. While many neural radial correlates were shared between clinical and digital meas- ures, the latter showed associations with brain regions Circle CV that clinical measures were unable to demonstrate. These velocity angular results indicate that remotely administered digital tests may yield higher functional specificity and provide poten- Double touch asynchrony tial novel information on functional ability compared with Pinching Test standard clinical measures. Of note, these correlations were observed in an early, relapsing cohort of PwMS, Pinching gap time ◂ Fig. 3  Spearman’s rank correlations adjustment for age, sex, and BMI. Statistically significant (q < 0.05) correlations between regional brain volume measured by MRI and measures of a the cognitive 9HPT Time domain, b upper extremity function domain, and c balance domain are shown for four planes (from left to right: cortical outer, inner, top, and subcortical). FDR correction was applied for each digital and c standard clinical measure separately to correct for multiple compari- Sway path sons. Higher values equate to better performance for digital measures SBT assessing number of correct responses and SFI 30 on the e-SDMT; trace accuracy and trace celerity on the Draw a Shape Test; and dou- ble touch asynchrony on the Pinching Test. In contrast, higher values Balance Scale equate to worse performance for digital measures assessing max. gap Total score Berg time between correct responses on the e-SDMT; CV linear, angular, Spearman’s rho and radial velocity on the Draw a Shape Test; pinching gap time on –0.6 –0.3 0.0 0.3 0.6 the Pinching Test; and sway path on the SBT. 9HPT Nine-Hole Peg Test, BMI body mass index, btw between, CV coefficient of variation, e-SDMT smartphone-based electronic Symbol Digit Modalities Test, FDR false discovery rate, SBT Static Balance Test, SDMT Symbol Digit Modalities Test, SFI speed fatigability index 13 Journal of Neurology Number of correct responses suggesting that digital measures hold potential to capture silent pathology [31]. If confirmed in future work, this Max. gap time btw correct responses e-SDMT would address a key unmet need in MS disease manage- SFI 30 ment [32]. As expected, both the e-SDMT and oral SDMT cor- Mean trace accuracy related with regions implicated in cognitive decline (ven- Mean trace celerity tricular expansion and smaller volume of the insula and cerebral white matter), information processing speed (deep Figure-of-8 CV linear velocity Draw a Shape Test gray matter), and working memory (superior temporal Circle CV angular velocity gyrus) [33–39]. Correlations were also observed with the Spiral CV radial velocity volume of the primary somatosensory cortex (postcentral gyrus) and visual processing areas (lateral occipital lobe) Double touch asynchrony [34, 40]. Similarly, the Draw a Shape Test, Pinching Test, Pinching gap time Pinching Test and 9HPT were associated with areas of visual process- ing (lateral occipital lobe, deep gray matter) and cogni- Number of successful pinches tive decline (ventricular expansion, cerebral white matter) Sway path SBT [34–36]. For the SBT, postural control (sway path) corre- Step power mean lated inversely with the paracentral lobule volume. This is consistent with findings from a functional imaging study, Step frequency mean 2MWT which suggested a role of the paracentral lobule in pro- Step frequency variance prioceptive processing [41]. In contrast, the Berg Balance Scale correlated more strongly with the deep gray matter Turn speed mean UTT volume. The different tasks involved in the two tests, the 9HPT distinctive sensorimotor aspects they capture, and the scor- ing—the SBT measures the total sway path (a measure Oral SDMT of static balance), while the Berg Balance Scale provides T25FW Clinical measures an overall score describing both static and dynamic bal- Berg Balance Scale ance—may explain these differences. The mild level of MS-related impairment together with the ceiling effect EDSS observed on the Berg Balance Scale may have contributed to the lack of correlations with the cerebellar volume. The Parcellation Whole Brain R2 former may also explain the lack of correlations found 0.00 0.35 for the sensor-based walking tests (UTT, 2MWT) or the T25FW [42]. With greater levels of impairment, correla- tions between the T25FW and thalamic volumes or upper Fig. 4  Bayesian ridge regression model with leave-one-out cross val- cervical cord area could be expected [43, 44]. idation for estimating the variance (R2) in the digital measures and Several correlations were only observed for the digital standard clinical measures that can be explained by volumetric MRI. measures, but not for the standard clinical measures. This Two models were applied. The first model (“Whole Brain”) included normalized brain volume and the three demographic variables: age, was most evident in the upper extremity function domain. sex, and body mass index (top row). The second model (“Parcella- Temporal and spatiotemporal measures derived from the tion”) included all 36 regional MRI regions and the same three demo- Draw a Shape Test such as the mean trace celerity and vari- graphic variables (bottom row). Compared with the standard clinical ability of drawing velocity while drawing round shapes spe- measures, the digital measures tend to have a larger R2 score in the Parcellation model vs. in the Whole Brain model, which may reflect cifically correlated with thalamic volume, which has been higher functional specificity. This is particularly evident on the Draw previously shown to be involved in information processing a Shape Test (mean trace celerity and spiral CV radial velocity) and [45]. These measures also specifically correlated with the on the Pinching Test (double touch asynchrony). 2MWT Two-Minute volume of the pons. Brainstem atrophy is reported in early Walk Test, 9HPT Nine-Hole Peg Test, btw between, CV coefficient of variation, EDSS Expanded Disability Status Scale, e-SDMT smart- stages of MS and a higher pontine lesion load has been phone-based electronic Symbol Digit Modalities Test, SBT Static associated with upper extremity tremor [7, 46]. In addition, Balance Test, SDMT Symbol Digit Modalities Test, SFI speed fatiga- the variability of linear drawing velocity on the figure-of-8 bility index, T25FW Timed 25-Foot Walk, UTT​U-Turn Test specifically correlated with the upper cervical cord area. This region has been previously linked to upper extremity dysfunction, particularly in more advanced or progressive disease where atrophy is more pronounced [44, 47]. The fact 13 Journal of Neurology that we observed this correlation in a mildly impaired cohort both cortical and subcortical structural networks [54, 55] highlights the potential higher sensitivity of Floodlight PoC and assess the relationship between longitudinal changes in digital measures. Correlations specific to the digital meas- digital measures and MRI outcomes to better characterize ures were also observed with the Pinching Test. Double the utility of sensor-based tests as prognostic biomarkers. touch asynchrony, that corresponds to an asynchronous con- Such biomarkers could be used for early identification of tact of the two fingers with the touchscreen while pinching a patients with silent progression at risk of future disability tomato, was associated with the volume of anterior cingulate accrual, and optimization of individual treatment strategies. cortex. No association was observed with the 9HPT time. This region is known to selectively modulate motor areas during visually coordinated tasks [48, 49]. This result is not Conclusions surprising considering that double touch asynchrony was specifically developed to assess the ability to perform finger In this exploratory post-hoc analysis, digital measures coordination tasks. obtained with the Floodlight PoC app correlated with nor- The notion that digital measures may offer higher func- malized volumes of distinct anatomical regions. While many tional specificity is also supported by the explained variance of the correlations were also observed with standard clinical analysis. A low R2 value in the Whole Brain model but a measures, some were only observed with the digital meas- high R2 value in the Parcellation model indicates that certain ures. In addition, the explained variance analysis may sug- regions contribute more than other regions to the variance gest a higher functional specificity for digital measures, in observed in the digital or standard clinical measures. This particular in the upper extremity function domain. These can be seen in the upper extremity function domain, where results indicate that digital measures, by leveraging sensor selected measures derived from the Draw a Shape and Pinch- technology, can probe multiple different neurologic domains ing Tests showed a larger increase in R2 when switching rather than just providing an overall assessment of func- from the Whole Brain to the Parcellation model compared tional ability. Thus, digital measures have the potential to with the 9HPT. This suggests that these digital measures complement standard clinical measures by providing a more have higher functional specificity than the 9HPT. A high detailed picture of MS and a more accessible assessment of R2 value in both models, on the other hand, indicates that functional ability. Larger, ongoing, and forthcoming studies the different MRI regions contribute equally to the observed will need to confirm these preliminary findings. variance. The cognitive domain with both the e-SDMT and Supplementary Information The online version contains supplemen- oral SDMT is a good example of this, with both showing tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00415-0​ 22-1​ 1494-0. comparable functional specificity. There are some limitations to this study. First, MRI data Acknowledgements We would like to thank all patients, their fami- were only collected for PwMS. Consequently, we were lies, and the investigators who participated in this trial. We would like to thank the following colleagues at F. Hoffmann-La Roche Ltd for unable to disentangle physiologic from pathologic effects. their contributions and support to the study: Jan Beckmann, Sandro Second, the relatively short study duration of 24 weeks did Fritz, Nicholas Pierce Heinemeier, Timothy Kilchenmann, Lito Kri- not allow us to perform longitudinal analyses or assess the ara, Grégoire Pointeau, Cedric Simillion, Jens Schjodt-Eriksen, Jörg relationship between the digital measures and disease pro- Sprengel, and Mattia Zanon. Writing and editorial assistance for this manuscript was provided by Terri Desmonds, PhD, from Articulate gression. Third, PwMS enrolled in this study mostly had Science, United Kingdom, and funded by F. Hoffmann-La Roche Ltd. mild disease with limited functional impairment, which may have weakened the correlations. Structural, or even func- Author contributions Design or conceptualization of the study: CB, tional, cortical reorganization, that helps to maintain func- ML. Analysis or interpretation of the data: MG, JSG, SPH, FD, LG, LC. Drafting and revising the manuscript for intellectual content: MG, tional ability in early stages of the disease despite structural JSG, SPH, FD, LM, LG, LC, FL, CB, XM, SLH, ML. damage to the brain [50–53], may have further weakened the observed correlations. Finally, spurious correlations can- Funding This research was funded by F. Hoffmann-La Roche Ltd, not be excluded. However, the observed effects remained Basel, Switzerland. largely statistically significant even when applying a more Data availability For up-to-date details on Roche's Global Policy on conservative correction for multiple testing (FDR correction Sharing of Clinical Study Information and how to request access to applied for all the possible combinations of the 36 anatomi- related clinical study documents, see here: https://​go.​roche.​com/​data_​ cal regions and 21 digital/clinical measures [36 × 21 con- shari​ng. Request for the data underlying this publication requires a detailed, hypothesis-driven, statistical analysis plan that is collabora- figuration]), suggesting that potential spurious correlations tively developed by the requestor and company subject matter experts. are negligible (Fig. SI8). Such requests should be directed to dbm.datarequest@roche.com for In the future, using data from larger, ongoing, and forth- consideration. Anonymized records for individual patients across more coming studies (CONSONANCE, NCT03523858; Flood- than one data source external to Roche cannot, and should not, be linked due to a potential increase in risk of patient re-identification. light™ MS—TONiC, ISRCTN11088592), we will explore 13 Journal of Neurology Declarations sclerosis: an MRI sub-regional volumetric analysis. Mult Scler 23:1194–1203. https://​doi.​org/​10.​1177/​13524​58516​674567 Conflicts of interest MG is a contractor for F. Hoffmann-La Roche 4. Schlaeger R, Papinutto N, Panara V et al (2014) Spinal cord gray Ltd. JSG over the past year has received grant/contract research sup- matter atrophy correlates with multiple sclerosis disability. Ann port from the National MS Society, Biogen, and Octave Biosciences. Neurol 76:568–580. https://​doi.​org/​10.​1002/​ana.​24241 She serves on a steering committee for a trial supported by Novartis. 5. Pagani E, Rocca MA, Gallo A et al (2005) Regional brain atrophy She has received honoraria for a non-promotional, educational activ- evolves differently in patients with multiple sclerosis according to ity for Sanofi Genzyme. She has received speaker fees from Alexion clinical phenotype. AJNR Am J Neuroradiol 26:341–346 and BMS and served on an advisory board for Genentech. SPH is a 6. Eshaghi A, Prados F, Brownlee WJ et al (2018) Deep gray matter contractor for F. Hoffmann-La Roche Ltd. FD was an employee of and volume loss drives disability worsening in multiple sclerosis. Ann is a shareholder in F. Hoffmann-La Roche Ltd; he is currently em- Neurol 83:210–222. https://​doi.​org/​10.​1002/​ana.​25145 ployed by Novartis Institutes for Biomedical Research. LM has noth- 7. Eshaghi A, Marinescu RV, Young AL et al (2018) Progression ing to disclose. LG is an employee of F. Hoffmann-La Roche Ltd. LC of regional grey matter atrophy in multiple sclerosis. Brain is an employee of and shareholder in F. Hoffmann-La Roche Ltd. FL 141:1665–1677. https://​doi.​org/​10.​1093/​brain/​awy088 is an employee of F. Hoffmann-La Roche Ltd. CB is a contractor for 8. Cordani C, Hidalgo de la Cruz M, Meani A et al (2020) MRI F. Hoffmann-La Roche Ltd. XM has received speaking honoraria and correlates of clinical disability and hand-motor performance in travel expenses for participation in scientific meetings; has been a multiple sclerosis phenotypes. Mult Scler 27:1205–1221. https://​ steering committee member of clinical trials; or participated in advi- doi.​org/​10.​1177/​13524​58520​958356 sory boards of clinical trials in the past years with Actelion, Alexion, 9. Artemiadis A, Anagnostouli M, Zalonis I et al (2018) Structural Bayer, Biogen, Celgene, EMD Serono, Genzyme, Immunic, Medday, MRI correlates of cognitive function in multiple sclerosis. Mult Merck, Mylan, Nervgen, Novartis, Roche, Sanofi-Genzyme, Teva, TG Scler Relat Disord 21:1–8. https://​doi.​org/​10.​1016/j.​msard.​2018.​ Therapeutics, Excemed, MSIF, and NMSS. SLH serves on the sci- 02.​003 entific advisory boards for Alector, Annexon, and Accure; is on the 10. Jakimovski D, Weinstock-Guttman B, Hagemeier J et al (2018) Board of Directors for Neurona Therapeutics; and has received travel Walking disability measures in multiple sclerosis patients: cor- reimbursement and writing assistance from F. Hoffmann-La Roche Ltd relations with MRI-derived global and microstructural damage. and Novartis for CD20-related meetings and presentations. ML is a J Neurol Sci 393:128–134. https://​doi.​org/​10.​1016/j.​jns.​2018.​08.​ consultant for F. Hoffmann-La Roche Ltd via Inovigate. 020 11. Mistri D, Cacciaguerra L, Storelli L et al (2022) The associa- Ethics approval The study was registered on clinicaltrials.gov tion between cognition and motor performance is beyond struc- (NCT02952911). The study was conducted in full conformance with tural damage in relapsing-remitting multiple sclerosis. J Neurol the ICH E6 guideline for Good Clinical Practice and the principles of 269:4213–4221. https://​doi.​org/​10.​1007/​s00415-​022-​11044-8 the Declaration of Helsinki. All participants provided informed consent 12. Sormani MP, De Stefano N, Giovannoni G et al (2019) Learning and ethical approval was obtained from the ethics committee of the ability correlates with brain atrophy and disability progression in Vall d’Hebron University Hospital, Barcelona, Spain (study number: RRMS. J Neurol Neurosurg Psychiatry 90:38–43. https://​doi.​org/​ PR(AG)300/2016) and the institutional review board of the University 10.​1136/​jnnp-​2018-​319129 of California San Francisco, San Francisco, CA, USA (reference num- 13. Montalban X, Graves J, Midaglia L et al (2022) A smartphone ber: 175728) prior to study initiation. sensor-based digital outcome assessment of multiple sclerosis. Mult Scler 28:654–664. https://​doi.​org/​10.​1177/​13524​58521​ 10285​61 Open Access This article is licensed under a Creative Commons Attri- 14. Midaglia L, Mulero P, Montalban X et al (2019) Adherence and bution 4.0 International License, which permits use, sharing, adapta- satisfaction of smartphone- and smartwatch-based remote active tion, distribution and reproduction in any medium or format, as long testing and passive monitoring in people with multiple sclerosis: as you give appropriate credit to the original author(s) and the source, nonrandomized interventional feasibility study. J Med Internet Res provide a link to the Creative Commons licence, and indicate if changes 21:e14863. https://​doi.​org/​10.​2196/​14863 were made. The images or other third party material in this article are 15. Polman CH, Reingold SC, Banwell B et al (2011) Diagnostic cri- included in the article's Creative Commons licence, unless indicated teria for multiple sclerosis: 2010 revisions to the McDonald cri- otherwise in a credit line to the material. If material is not included in teria. Ann Neurol 69:292–302. https://​doi.​org/​10.​1002/​ana.​22366 the article's Creative Commons licence and your intended use is not 16. Kurtzke JF (1983) Rating neurologic impairment in multiple scle- permitted by statutory regulation or exceeds the permitted use, you will rosis: an Expanded Disability Status Scale (EDSS). Neurology need to obtain permission directly from the copyright holder. To view a 33:1444–1452. https://​doi.​org/​10.​1212/​wnl.​33.​11.​1444 copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. 17. Jain S, Sima DM, Ribbens A et al (2015) Automatic segmenta- tion and volumetry of multiple sclerosis brain lesions from MR images. Neuroimage Clin 8:367–375. https://​doi.​org/​10.​1016/j.​ nicl.​2015.​05.​003 References 18. Rakić M, Vercruyssen S, Van Eyndhoven S et al (2021) icobrain ms 5.1: combining unsupervised and supervised approaches for 1. Reich DS, Lucchinetti CF, Calabresi PA (2018) Multiple sclerosis. improving the detection of multiple sclerosis lesions. Neuroimage N Engl J Med 378:169–180. https://d​ oi.o​ rg/1​ 0.1​ 056/N ​ EJMra​ 1401​ Clin 31:102707. https://​doi.​org/​10.​1016/j.​nicl.​2021.​102707 483 19. Sastre-Garriga J, Pareto D, Battaglini M et al (2020) MAGNIMS 2. Steenwijk MD, Geurts JJ, Daams M et al (2016) Cortical atro- consensus recommendations on the use of brain and spinal cord phy patterns in multiple sclerosis are non-random and clinically atrophy measures in clinical practice. Nat Rev Neurol 16:171–182. relevant. Brain 139(Pt 1):115–126. https://​doi.​org/​10.​1093/​brain/​ https://​doi.​org/​10.​1038/​s41582-​020-​0314-x awv337 20. U.S. Food and Drug Administration (2019) Substantial equiva- 3. D’Ambrosio A, Pagani E, Riccitelli GC et al (2017) Cerebellar lence determination: icobrain. Regulation Number: 21 CFR contribution to motor and cognitive performance in multiple 892.2050. https://​www.​acces​sdata.​fda.​gov/​cdrh_​docs/​pdf19/​ K1921​30.​pdf. Accessed 13 Oct 2022 13 Journal of Neurology 21. Struyfs H, Sima DM, Wittens M et al (2020) Automated MRI performance in multiple sclerosis using voxel-based morphome- volumetry as a diagnostic tool for Alzheimer’s disease: validation try. Front Neurol 9:920. https://d​ oi.o​ rg/1​ 0.3​ 389/f​ neur.2​ 018.0​ 0920 of icobrain dm. Neuroimage Clin 26:102243. https://​doi.​org/​10.​ 38. Zhang X, Zhang F, Huang D et al (2016) Contribution of gray and 1016/j.​nicl.​2020.​102243 white matter abnormalities to cognitive impairment in multiple 22. Steenwijk MD, Amiri H, Schoonheim MM et al (2017) Agree- sclerosis. Int J Mol Sci 18:46. https://​doi.​org/​10.​3390/​ijms1​80100​46 ment of MSmetrix with established methods for measuring 39. Batista S, Zivadinov R, Hoogs M et al (2012) Basal ganglia, thala- cross-sectional and longitudinal brain atrophy. Neuroimage Clin mus and neocortical atrophy predicting slowed cognitive process- 15:843–853. https://​doi.​org/​10.​1016/j.​nicl.​2017.​06.​034 ing in multiple sclerosis. J Neurol 259:139–146. https://​doi.​org/​ 23. Liu Y, Lukas C, Steenwijk M et al (2016) Multicenter valida- 10.​1007/​s00415-​011-​6147-1 tion of mean upper cervical cord area measurements from head 40. Riccitelli GC, Pagani E, Rodegher M et al (2019) Imaging patterns 3D T1-weighted MR imaging in patients with multiple sclerosis. of gray and white matter abnormalities associated with PASAT AJNR Am J Neuroradiol 37:749–754. https://​doi.​org/​10.​3174/​ and SDMT performance in relapsing-remitting multiple sclero- ajnr.​A4635 sis. Mult Scler 25:204–216. https://​doi.​org/​10.​1177/​13524​58517​ 24. Liu Z, Yaldizli Ö, Pardini M et al (2015) Cervical cord area meas- 743091 urement using volumetric brain magnetic resonance imaging in 41. Goossens N, Janssens L, Caeyenberghs K, Albouy G, Brumagne multiple sclerosis. Mult Scler Relat Disord 4:52–57. https://​doi.​ S (2019) Differences in brain processing of proprioception related org/​10.​1016/j.​msard.​2014.​11.​004 to postural control in patients with recurrent non-specific low back 25. Van Hecke W, Billiet T, Pedrosa de Barros N et al (2019) Evalu- pain and healthy controls. Neuroimage Clin 23:101881. https://​ ation of spinal cord atrophy in different MS subtypes using doi.​org/​10.​1016/j.​nicl.​2019.​101881 icospine. In: Poster presented at: 3­ 7th Congress of the European 42. Maghzi AH, Revirajan N, Julian LJ et al (2014) Magnetic reso- Committee for Treatment and Research in Multiple Sclerosis; nance imaging correlates of clinical outcomes in early multiple September 11–13; Stockholm, Sweden sclerosis. Mult Scler Relat Disord 3:720–727. https://​doi.​org/​10.​ 26. Pedrosa de Barros N, Billiet T, Denissen S et al (2020) Assessing 1016/j.​msard.​2014.​07.​003 cervical spinal cord atrophy in T1-weighted brain scans using ico- 43. Bergsland N, Benedict RHB, Dwyer MG et al (2021) Thalamic brain. In: Poster presented at: American Society of Neuroimaging nuclei volumes and their relationships to neuroperformance in 43rd Annual Meeting; March 5–7; Atlanta, GA, USA multiple sclerosis: a cross-sectional structural MRI study. J Magn 27. De Leener B, Lévy S, Dupont SM et al (2017) SCT: Spinal Cord Reson Imaging 53:731–739. https://​doi.​org/​10.​1002/​jmri.​27389 Toolbox, an open-source software for processing spinal cord MRI 44. Daams M, Weiler F, Steenwijk MD et al (2014) Mean upper cervi- data. Neuroimage Clin 145:24–43. https://d​ oi.o​ rg/1​ 0.1​ 016/j.n​ euro​ cal cord area (MUCCA) measurement in long-standing multiple image.​2016.​10.​009 sclerosis: relation to brain findings and clinical disability. Mult 28. Huber PJ (1981) Robust statistics. Wiley, New York Scler 20:1860–1865. https://​doi.​org/​10.​1177/​13524​58514​533399 29. Benjamini Y, Hochberg Y (1995) Controlling the false discovery 45. Merchant H, Harrington DL, Meck WH (2013) Neural basis of the rate: a practical and powerful approach to multiple testing. J R perception and estimation of time. Annu Rev Neurosci 36:313– Statist Soc B 57:289–300. https://​doi.​org/​10.​1111/j.​2517-​6161.​ 336. https://​doi.​org/​10.​1146/​annur​ev-​neuro-​062012-​170349 1995.​tb020​31.x 46. Feys P, Maes F, Nuttin B et al (2005) Relationship between mul- 30. MacKay DJC (1992) Bayesian interpolation. Neural Comput tiple sclerosis intention tremor severity and lesion load in the 4:415–447. https://​doi.​org/​10.​1162/​neco.​1992.4.​3.​415 brainstem. NeuroReport 16:1379–1382. https://​doi.​org/​10.​1097/​ 31. University of California San Francisco MS-EPIC Team, Cree BA, 01.​wnr.​00001​76521.​26971.​58 Hollenbach JA et al (2019) Silent progression in disease activity- 47. Rocca MA, Horsfield MA, Sala S et al (2011) A multicenter free relapsing multiple sclerosis. Ann Neurol 85:653–666. https://​ assessment of cervical cord atrophy among MS clinical pheno- doi.​org/​10.​1002/​ana.​25463 types. Neurology 76:2096–2102. https://​doi.​org/​10.​1212/​WNL.​ 32. Members of the MS in the 21st Century Steering Group, Rieck- 0b013​e3182​1f46b8 mann P, Centonze D et al (2018) Unmet needs, burden of treat- 48. Asemi A, Ramaseshan K, Burgess A, Diwadkar VA, Bressler SL ment, and patient engagement in multiple sclerosis: a combined (2015) Dorsal anterior cingulate cortex modulates supplementary perspective from the MS in the 21st Century Steering Group. Mult motor area in coordinated unimanual motor behavior. Front Hum Scler Relat Disord 19:153–160. https://​doi.​org/​10.​1016/j.​msard.​ Neurosci 9:309. https://​doi.​org/​10.​3389/​fnhum.​2015.​00309 2017.​11.​013 49. Paus T, Petrides M, Evans AC, Meyer E (1993) Role of the human 33. Achiron A, Chapman J, Tal S, Bercovich E, Gil H (2013) Superior anterior cingulate cortex in the control of oculomotor, manual, temporal gyrus thickness correlates with cognitive performance and speech responses: a positron emission tomography study. J in multiple sclerosis. Brain Struct Funct 218:943–950. https://d​ oi.​ Neurophysiol 70:453–469. https://​doi.​org/​10.​1152/​jn.​1993.​70.2.​ org/​10.​1007/​s00429-​012-​0440-3 453 34. Bisecco A, Stamenova S, Caiazzo G et al (2018) Attention and 50. Pantano P, Mainero C, Caramia F (2006) Functional brain reor- processing speed performance in multiple sclerosis is mostly ganization in multiple sclerosis: evidence from fMRI studies. J related to thalamic volume. Brain Imaging Behav 12:20–28. Neuroimaging 16:104–114. https://​doi.​org/​10.​1111/j.​1552-​6569.​ https://​doi.​org/​10.​1007/​s11682-​016-​9667-6 2006.​00029.x 35. Carmichael OT, Kuller LH, Lopez OL et al (2007) Cerebral 51. Rocca MA, Falini A, Colombo B, Scotti G, Comi G, Filippi ventricular changes associated with transitions between normal M (2002) Adaptive functional changes in the cerebral cortex cognitive function, mild cognitive impairment, and dementia. of patients with nondisabling multiple sclerosis correlate with Alzheimer Dis Assoc Disord 21:14–24. https://​doi.​org/​10.​1097/​ the extent of brain structural damage. Ann Neurol 51:330–339. WAD.​0b013​e3180​32d2b1 https://​doi.​org/​10.​1002/​ana.​10120 36. Engl C, Tiemann L, Grahl S et al (2020) Cognitive impairment in 52. Fleischer V, Gröger A, Koirala N et al (2017) Increased structural early MS: contribution of white matter lesions, deep grey matter white and grey matter network connectivity compensates for func- atrophy, and cortical atrophy. J Neurol 267:2307–2318. https://​ tional decline in early multiple sclerosis. Mult Scler 23:432–441. doi.​org/​10.​1007/​s00415-​020-​09841-0 https://​doi.​org/​10.​1177/​13524​58516​651503 37. Matías-Guiu JA, Cortés-Martínez A, Montero P et al (2018) 53. Llufriu S, Martinez-Heras E, Solana E et al (2017) Structural Identification of cortical and subcortical correlates of cognitive networks involved in attention and executive functions in multiple 13 Journal of Neurology sclerosis. Neuroimage Clin 13:288–296. https://d​ oi.o​ rg/1​ 0.1​ 016/j.​ 55. Pagani E, Rocca MA, De Meo E et al (2020) Structural connec- nicl.​2016.​11.​026 tivity in multiple sclerosis and modeling of disconnection. Mult 54. Colato E, Stutters J, Tur C et al (2021) Predicting disability pro- Scler 26:220–232. https://​doi.​org/​10.​1177/​13524​58518​820759 gression and cognitive worsening in multiple sclerosis using patterns of grey matter volumes. J Neurol Neurosurg Psychiatry 92:995–1006. https://​doi.​org/​10.​1136/​jnnp-​2020-​325610 13