Hybrid Model for Freezing of Gait Detection in Parkinson’s Disease: Integrating Manual Features and Deep Learning

Freezing of gait (FoG) is a common and debilitating symptom of Parkinson’s disease (PD), and its early detection is crucial for timely intervention and improved patient outcomes. Traditional approaches often rely on manually extracted gait features, which may overlook subtle yet clinically significant patterns. This study presents a hybrid model that integrates manual gait feature extraction with deep learning-based analysis using a 3D-ResNext model to enhance the detection of FoG. In the first stage, both manual and deep learning features are extracted, and in the second stage, these features are fed into a long short-term memory (LSTM) network to classify gait patterns into walk, Pre-FoG, and FoG. The model was evaluated using publicly available Daphnet dataset, which comprises gait data from ten PD patients, eight of whom exhibited FoG events. Experimental analysis with varying segment lengths and Pre-FoG periods revealed that the optimal performance was achieved with a 1-s segment length and a 5-s Pre-FoG window. The proposed hybrid model achieved an accuracy of 98.87%, specificity of 96.67%, F1-score of 92.29%, and a loss value of 0.0597. These results demonstrate that the proposed approach outperforms existing state-of-the-art models and offers a robust solution for accurate FoG classification and PD prediction.