Personalized prediction system for early prediction of freezing of gait in Parkinson's disease using explainable AI

Abstract Freezing of gait (FOG) is a common symptom of Parkinson’s disease (PD), characterized by sudden and temporary episodes of immobility, often resulting in falls and reduced quality of life. Early and accurate prediction of FOG can greatly improve patient outcomes by allowing timely intervention and tailored treatment strategies. This article presents a personalized system for the early prediction of gait freezing in PD patients using explainable artificial intelligence (XAI). The proposed system follows a subject-dependent approach, training models specifically for individual patients to enhance prediction accuracy. It utilized multiple explainability techniques to ensure transparency in the decision-making process, achieving an average accuracy of 98.67 ± 0.75% using the random forest (RF) model and a latency of 75.0 ± 31.1 ms across six patients. Feature importance analysis, including SHapley Additive exPlanations (SHAP) and local interpretable model-agnostic explanation (LIME) plots, revealed significant influences of features like the maximum value of the x-axis, the maximum of the accelerometer signal from the x-axis and the standard deviation of the gyroscope signal from the y-axis in classifying gait states.