A Vision-Based Approach for the Automated Evaluation of the MoCA Clock-Drawing Test Using a YOLO Neural Network

This paper presents an analysis of the clock-drawing test from the Montreal Cognitive Assessment (MoCA), using computer vision and deep learning techniques to perform an evaluation based on the Rouleau scale. The proposed system consists of 3 components: 1) computer vision techniques such as convex hull and morphological operations to evaluate the contour; 2) a combination of computer vision and the YOLO-v8 algorithm to assess the presence and correct positioning of numbers; and 3) an analysis of the hands using the Hough Line Transform. The methodology was evaluated using 29 samples from individuals with Parkinson's disease, and the results were benchmarked against assessments from two human evaluators, one of whom is a MoCA-certified examiner. The proposed approach achieved an inter-rater correlation of 0.915 with the certified examiner, highlighting the system's efficacy in analyzing clock-drawing tasks and potentially classifying cognitive impairment.