The clinical manifestations of Parkinson’s disease (PD) are characterized by heterogeneity in age at onset, disease duration, rate of progression, and the constellation of motor versus non-motor features. There is an unmet need for the characterization of distinct disease subtypes as well as improved, individualized predictions of the disease course. We used unsupervised and supervised machine learning methods on comprehensive, longitudinal clinical data from the Parkinson’s Disease Progression Marker Initiative (n = 294 cases) to identify patient subtypes and to predict disease progression. The resulting models were validated in an independent, clinically well-characterized cohort from the Parkinson’s Disease Biomarker Program (n = 263 cases). Our analysis distinguished three distinct disease subtypes with highly predictable progression rates, corresponding to slow, moderate, and fast disease progression. We achieved highly accurate projections of disease progression 5 years after initial diagnosis with an average area under the curve (AUC) of 0.92 (95% CI: 0.95 ± 0.01) for the slower progressing group (PDvec1), 0.87 ± 0.03 for moderate progressors, and 0.95 ± 0.02 for the fast-progressing group (PDvec3). We identified serum neurofilament light as a significant indicator of fast disease progression among other key biomarkers of interest. We replicated these findings in an independent cohort, released the analytical code, and developed models in an open science manner. Our data-driven study provides insights to deconstruct PD heterogeneity. This approach could have immediate implications for clinical trials by improving the detection of significant clinical outcomes. We anticipate that machine learning models will improve patient counseling, clinical trial design, and ultimately individualized patient care.

This study examined the effectiveness of a v ision-based framework for m ultiple s clerosis (MS) and Parkinson's disease (PD) gait dysfunction prediction. We collected gait video data from multi-view digital cameras during self-paced walking from MS, PD patients and age, weight, height and gender-matched healthy older adults (HOA). We then extracted characteristic 3D joint keypoints from the collected videos. In this work, we proposed a data-driven methodology to classify strides in persons with MS (PwMS), persons with PD (PwPD) and HOA that may generalize across different walking tasks and subjects. We presented a comprehensive quantitative comparison of 16 diverse traditional machine and deep learning (DL) algorithms. When generalizing from comfortable walking (W) to walking-while-talking (WT), multi-scale residual neural network achieved perfect accuracy and AUC for classifying individuals with a given gait disorder; for subject generalization in W trials, residual neural network resulted in the highest accuracy and AUC of $\mathbf {78.1\%}$ and 0.87 (resp.), and 1D convolutional neural network (CNN) had highest accuracy of $\mathbf {75\%}$ in WT trials. Finally, when generalizing over new subjects in different tasks, again 1D CNN had the top classification accuracy and AUC of $\mathbf {79.3\%}$ and 0.93 (resp.). This work is the first attempt to apply and demonstrate the potential of DL with a multi-view digital camera-based gait analysis framework for neurological gait dysfunction prediction. This study suggests the viability of inexpensive vision-based systems for diagnosing certain neurological disorders.

Objective: Multiple Sclerosis (MS) is a neurological condition which widely affects people 50-60 years of age. While clinical presentations of MS are highly heterogeneous, mobility limitations are one of the most frequent symptoms. This study examines a machine learning (ML) framework for identifying MS through spatiotemporal and kinetic gait features. Methods: In this study, gait data during self-paced walking on an instrumented treadmill from 20 persons with MS and 20 age, weight, height, and gender-matched healthy older adults (HOA) were obtained. We explored two strategies to normalize data and minimize dependence on subject demographics; size-normalization (standard body size-based normalization) and regress-normalization (regression-based normalization using scaling factors derived by regressing gait features on multiple subject demographics); and proposed an ML based methodology to classify individual strides of older persons with MS (PwMS) from healthy controls. We generalized both across different walking tasks and subjects. Results: We observed that regress-normalization improved the accuracy of identifying pathological gait using ML when compared to size-normalization. When generalizing from comfortable walking to walking while talking, gradient boosting machine achieved the optimal subject classification accuracy and AUC of 94.3 and 1.0, respectively and for subject generalization, a multilayer perceptron resulted in the best accuracy and AUC of 80 % and 0.86, respectively, both with regression-normalized data. Conclusion: The integration of gait data and ML may provide a viable patient-centric approach to aid clinicians in monitoring MS. Significance: The results of this study have future implications for the way regression normalized gait features may be clinically used to design ML-based disease prediction strategies and monitor disease progression in PwMS.

Objective: Multiple sclerosis (MS) is a chronic neurological condition of the central nervous system leading to various physical, mental and psychiatric complexities. Mobility limitations are amongst the most frequent and early markers of MS. We evaluated the effectiveness of a DeepMS2G ( deep learning (DL) for MS differentiation using m ulti- s tride dynamics in g ait) framework, which is a DL-based methodology to classify multi-stride sequences of persons with MS (PwMS) from healthy controls (HC), in order to generalize over newer walking tasks and subjects. Methods: We collected single-task Walking and dual-task Walking-while-Talking gait data using an instrumented treadmill from a balanced collection of 20 HC and 20 PwMS. We utilized domain knowledge-based spatiotemporal and kinetic gait features along with two normalization schemes, namely standard size-based and multiple regression normalization strategies. To differentiate between multi-stride sequences of HC and PwMS, we compared 16 traditional machine learning and DL algorithms. Further, we studied the interpretability of our highest-performing models; and discussed the association between the lower extremity function of participants and our model predictions. Results: We observed that residual neural network (ResNet) based models with regression-based normalization were the top performers across both task and subject generalization classification designs. Considering regression-based normalization, a multi-scale ResNet attained a subject classification accuracy and F $_{1}$ -score of 1.0 when generalizing from single-task Walking to dual-task Walking-while-Talking ; and a ResNet resulted in the top subject-wise accuracy and F $_{1}$ of 0.83 and 0.81 (resp.), when generalizing over unseen participants. Conclusion: We used advanced DL and dynamics across domain knowledge-based spatiotemporal and kinetic gait parameters to successfully classify MS gait across distinct walking trials and unseen participants. Significance: Our proposed DL algorithms might contribute to efforts to automate MS diagnoses.

Time series forecasting plays a crucial role in decision-making across various domains, but it presents significant challenges. Recent studies have explored image-driven approaches using computer vision models to address these challenges, often employing lineplots as the visual representation of time series data. In this paper, we propose a novel approach that uses time-frequency spectrograms as the visual representation of time series data. We introduce the use of a vision transformer for multimodal learning, showcasing the advantages of our approach across diverse datasets from different domains. To evaluate its effectiveness, we compare our method against statistical baselines (EMA and ARIMA), a state-of-the-art deep learning-based approach (DeepAR), other visual representations of time series data (lineplot images), and an ablation study on using only the time series as input. Our experiments demonstrate the benefits of utilizing spectrograms as a visual representation for time series data, along with the advantages of employing a vision transformer for simultaneous learning in both the time and frequency domains.

Wearable sensors provide a tool for at-home monitoring of motor impairment progression in neurological conditions such as Parkinson’s disease (PD). This study examined the ability of deep learning approaches to grade the motor impairment severity in a modified version of the Movement Disorders Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) using low-cost wearable sensors. We hypothesized that expanding training datasets with motion data from healthy older adults (HOAs) and initializing classifiers with weights learned from unsupervised pre-training would lead to an improvement in performance when classifying lower vs. higher motor impairment relative to a baseline deep learning model (XceptionTime). This study evaluated the change in classification performance after using expanded training datasets with HOAs and transferring weights from unsupervised pre-training compared to a baseline deep learning model (XceptionTime) using both upper extremity (finger tapping, hand movements, and pronation–supination movements of the hands) and lower extremity (toe tapping and leg agility) tasks consistent with the MDS-UPDRS. Overall, we found a 12.2% improvement in accuracy after expanding the training dataset and pre-training using max-vote inference on hand movement tasks. Moreover, we found that the classification performance improves for every task except toe tapping after the addition of HOA training data. These findings suggest that learning from HOA motion data can implicitly improve the representations of PD motion data for the purposes of motor impairment classification. Further, our results suggest that unsupervised pre-training can improve the performance of motor impairment classifiers without any additional annotated PD data, which may provide a viable solution for a widely deployable telemedicine solution.