An Intelligent Facial Palsy Diagnostic System Based on Acupoint Identification

Abstract

The methods for clinically diagnosing facial paralysis require doctors to possess a high degree of experience and specialized knowledge, often involving subjectivity. However, due to the uneven distribution of medical resources, many facial paralysis patients are unable to receive timely and accurate diagnosis and treatment. Traditional computer-assisted methods place high demands on hardware equipment and lack sufficient intelligence. With the continuous advancement of artificial intelligence, researchers have actively explored intelligent methods for facial paralysis detection. These methods mainly focus on extracting facial features and making judgments based on facial asymmetry, but they struggle to provide a scientific quantitative analysis of the severity of facial paralysis. This study is based on the lightweight network—MobileNetV2. By performing facial detection and processing on input images, it successfully identifies three groups of acupoints related to facial paralysis and conducts quantitative analysis based on this identification. Simultaneously, we have improved the network by constructing a two-stage network similar to object detection and regression, and optimizing the loss function. In the end, we compared the improved model with other mainstream frameworks through experiments. The results demonstrate that our proposed model achieves significant effectiveness in acupoint recognition and maintains low error in quantitative analysis.