Automated Assessment of Upper Extremity Function with the Modified Mallet Score Using Single-Plane Smartphone Videos

The Modified Mallet Score (MMS) is widely used to assess upper limb function but requires evaluation by experienced clinicians. This study automated MMS assessments using smartphone videos, artificial intelligence (AI), and new algorithms. A total of 125 videos covering all MMS grades were recorded from four neurotypical participants. For all recordings, an expert physician provided manual scores as the ground truth. The OpenPose BODY25 model extracted body keypoint data, which were used to calculate joint angles for an automated scoring algorithm. The algorithm's scores were compared to the ground truth and expert manual scoring. High accuracy was achieved for the global abduction, hand-to-neck, hand-on-spine, and hand-to-mouth movements, with Pearson correlation coefficients (PCCs) > 0.9 and a low root mean square error (RMSE). Although slightly less accurate for global external rotation, the algorithm still showed strong agreement. This study demonstrates the potential of using AI and smartphone videos for reliable, remote upper limb assessments.