Smartphone- and Smartwatch-Based Remote Characterisation of Ambulation in Multiple Sclerosis During the Two-Minute Walk Test

Abstract

Leveraging consumer technology such as smartphone and smartwatch devices to objectively assess people with multiple sclerosis (PwMS) remotely could capture unique aspects of disease progression. This study explores the feasibility of assessing PwMS and Healthy Control's (HC) physical function by characterising gait-related features, which can be modelled using machine learning (ML) techniques to correctly distinguish subgroups of PwMS from healthy controls. A total of 97 subjects (24 HC subjects, 52 mildly disabled (PwMSmild, EDSS [0-3]) and 21 moderately disabled (PwMSmod, EDSS [3.5-5.5]) contributed data which was recorded from a Two-Minute Walk Test (2MWT) performed out-of-clinic and daily over a 24-week period. Signal-based features relating to movement were extracted from sensors in smartphone and smartwatch devices. A large number of features (n=156) showed fair-to-strong (R>0.3) correlations with clinical outcomes. LASSO feature selection was applied to select and rank subsets of features used for dichotomous classification between subject groups, which were compared using Logistic Regression (LR), Support Vector Machines (SVM) and Random Forest (RF) models. Classifications of subject types were compared using data obtained from smartphone, smartwatch and the fusion of features from both devices. Models built on smartphone features alone achieved the highest classification performance, indicating that accurate and remote measurement of the ambulatory characteristics of HC and PwMS can be achieved with only one device. It was observed however that smartphonebased performance is affected by inconsistent placement location (running belt versus pocket). Results show that PwMSmod can be distinguished from HC subjects (Acc. 82.2 ± 2.9%, Sen. 80.1

Predicting Multiple Sclerosis From Gait Dynamics Using an Instrumented Treadmill: A Machine Learning Approach

“…Further, the study is limited in examining only force data and not exploring any tree-based ML algorithms. A recent study used smartphone and smartwatch sensors data and ML to distinguish among healthy controls, mildly (PwMS mild ) and moderately (PwMS mod ) disabled PwMS during a two-minute walk test [23]. Although this work investigates three wellknown algorithms, namely, LR, SVM and RF to achieve the best accuracy of 82% differentiating PwMS mod from HOA and from PwMS mild and 66% identifying PwMS mild from HOA; the analysis on boosting algorithms, which have known to outperform RF in most applications, is missing.…”

“…Although this work investigates three wellknown algorithms, namely, LR, SVM and RF to achieve the best accuracy of 82% differentiating PwMS mod from HOA and from PwMS mild and 66% identifying PwMS mild from HOA; the analysis on boosting algorithms, which have known to outperform RF in most applications, is missing. Moreover, our study utilizes up to 75 s of data for analysis, as compared to the longer data sample of two-minute walk in [23]. Another recent work analyzed a long short-term memory approach to classify fall risk in PwMS using accelerometers [37].…”

“…Exploring hidden Markov and recurrent neural network predictive models by using tensors of independent strides will be vital to gauge the temporal component present in the continuous gait data. Future work is needed to identify prospective fall risk in MS subjects and assess the performance of our approach with remotely acquired gait data [23] and wearable sensors [37]. Further, observed correlations of gait parameters with disability may help identify older PwMS advancing into sudden worsening, which may provide improved personalized care, and merits future investigation.…”

“…Supervised ML methodologies such as random forest and artificial neural networks have already been used in human gait analysis across other neurological populations [20], [21]. A few prior works have explored ML to classify MS using gait data [22], [23]. However, to the best of our knowledge, there is no study utilizing ML on spatiotemporal and kinetic gait characteristics for MS prediction.…”

Human Gait Analysis in Neurodegenerative Diseases: A Review

“…Healthy controls). SVM has been also used and compared to RF in [121] for classifying people with moderate MS (MS-mod), people with mild MS (MS-mild) and healthy controls. SVM performed best at distinguishing healthy controls from subjects with MS-mild and MS-mod, whereas the RF was marginally better at separating MS-mild vs. MS-mod.…”