Objective: To train and validate a machine learning model (MLM) algorithm utilizing the inertial and sEMG datasets for spasticity assessment according to MAS classification and determine the trained and validated MLM algorithm’s prediction performance in predicting ambiguous spasticity datasets.

Background: The MLM has garnered popularity in rehabilitation, ranging from developing algorithms in outcome prediction, prognostication, and training artificial intelligence to measure the rehabilitation interventions’ effectiveness. High-quality data plays a critical role in algorithm development. Biases and variances have been described as one of the challenges in ensuring high-quality data. However, limited studies have explored factors that may influence the MLM algorithm performance in predicting spasticity severity level.

Method: It is a prospective cross-sectional observational study. Forty-seven persons diagnosed with central nervous system pathology who underwent inpatient and outpatient rehabilitation clinic that fulfilled the inclusion and exclusion criteria were recruited. The Modified Ashworth Scale (MAS) classification of elbow flexors was evaluated by two experienced assessors, a rehabilitation physician, and a physiotherapist. Four biomechanical properties of spasticity were obtained using off-the-shelf wearable sensors. The data were analyzed individually, and ambiguous datasets for each MAS level classification were isolated and separated. The acceptable data of angle [degree], resistance force [newton], and surface electromyography [voltage] were included for the training and validating machine learning-based model (MLM) of MAS score. The trained and validated MLM algorithm was later deployed to predict the ambiguous spasticity datasets’ MAS score.

Results: A series of MLM were applied, including Support Vector Machine, Decision Tree, and Random Forest. The validated MLM’s performance accuracy was 96%, 52%, and 72%, respectively. The validated MLM accuracy performance level predicting ambiguous spasticity datasets reduces to 20%, 23%, and 23%, respectively. The low prediction performance are  MAS 0, MAS1+, MAS 2, and MAS 3.

Conclusion: This study elucidates data biases and variances that limits the MLM algorithm in predicting spasticity data.   Background diseases, pathophysiological and anatomical factors  have to be considered to interpret individual data.

« Back to MDS Virtual Congress 2021

MDS Abstracts - https://www.mdsabstracts.org/abstract/elucidating-factors-influencing-machine-learning-algorithm-prediction-in-spasticity-assessment-a-prospective-cross-sectional-observational-study/