Early Recognition of Facial Paralysis for Rehabilitation of Stroke Patients Using Visual Perception and AI-Assisted Analysis

Facial paralysis is one of the symptoms of neurological diseases such as stroke and Bell's paralysis. It causes a partial loss of facial muscle control, resulting in some facial asymmetry. The traditional diagnosis needs experienced professional doctors. We propose a vision-based facial image acquisition and auxiliary diagnosis system for patients with facial paralysis. The acquisition system based on an industrial camera and array microphone collects high-quality patient images, which are preprocessed and input into the auxiliary diagnostic system for image analysis. For auxiliary diagnosis, an image classification algorithm based on fusion feature was adopted to extract Histogram of Oriented Gradients (HoG) features, facial landmark features, and learning features. The output of auxiliary diagnosis includes normal, left weakness, and right weakness. The diagnostic effect was verified on a self-made dataset. The proposed auxiliary diagnosis algorithm achieves a classification accuracy of 96.1%, recall of 91.6%, precision of 96.2%, and F2 of 92.5%. The system is suitable for auxiliary diagnosis in hospitals and early self-examination at home for the rehabilitation of stroke people. It has higher efficiency than traditional methods.