Fig. 2

Study workflow. First, we gathered gait videos of patients with Parkinson’s disease, which were recorded using a consumer-grade video camera. Ten-second clips were then cropped from these videos, and 16 static frames were extracted at equal intervals. After preprocessing of the frames, they were inserted into the ECO-Lite convolutional neural network (CNN) model for the prediction of the Unified Parkinson’s Disease Rating Scale (UPDRS) score. The model was trained using training and validation datasets. Finally, we compared the UPDRS scores rated by neurologists and those predicted by the CNN model using the test dataset