Development and Validation of a Machine Learning-based Model to Predict a High-risk Group for Falls Using Multi-sensor Signals in Stroke Patients

Objective: The primary objective is to develop and validate a machine learning-based model that uses multi-sensor (EMG) signals to identify stroke patients at high risk of falls. This model aims to improve on traditional fall risk assessments which rely heavily on physical assessments and patient history. Study Design: This is a prospective, multicenter, open-label, confirmatory clinical trial. It involves collecting EMG data from stroke patients and applying machine learning techniques to predict fall risk. The study will compare the predictive accuracy of the machine learning model against conventional fall risk assessment tools. Methods: 1. Participants: • Sample Size: 80 stroke patients and 10 healthy adults to establish baseline EMG readings. 2. Interventions: • Participants will undergo EMG signal collection from key lower limb muscles while performing standardized movements. 3. Outcome Measures: * Primary Outcome: Sensitivity and specificity of the machine learning model in predicting high-risk fall patients. * Secondary Outcomes: Comparison of the machine learning model's predictive performance with traditional fall risk assessment tools (e.g., Berg Balance Scale). Data Collection: * EMG sensors will be attached to the patients' muscles of the lower limbs. Sensors will record muscle activity during movement, which will then be analyzed using the machine learning model. * The predictive model will be trained using features extracted from the EMG signals, and its performance will be validated against actual fall incidents reported during the follow-up period. Statistical Analysis: * The machine learning model's efficacy will be measured through its sensitivity (ability to correctly identify high-risk patients) and specificity (ability to correctly identify low-risk patients). * Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) statistics will be used to assess model performance.