Diagnosing facial synkinesis using artificial intelligence to advance facial palsy care.

Facial palsy (FP) can lead to significant psychological and physical burdens such as facial synkinesis. This involuntary simultaneous movement of facial musculature remains challenging to diagnose and treat. This study aimed to develop a cost-effective, rapid, and accurate artificial intelligence (AI)-based algorithm to screen FP patients for facial synkinesis. Data from 70 FP patients were collected at the University Hospital Regensburg and compared to healthy controls from an online platform. The standardized patient image series included 9 images, of which 3 were used to train the algorithm. The control images were single images. A total of 385 images were used to train and evaluate a convolutional neural network (CNN). The dataset was divided into training (285 images), validation (29 images), and test sets (71 images). The model was trained over 18 epochs. A web application was developed for practical use. The model achieved an accuracy of 98.6% on the test set, correctly identifying 31 of 32 synkinesis cases and all 39 images of healthy individuals. Performance metrics included an F1-score of 98.4%, precision of 100%, and recall of 96.9%. The web application allowed for image upload and rapid synkinesis prediction. The CNN-based model demonstrated high accuracy in detecting synkinesis in FP patients, offering potential for improved diagnostic accuracy and expedited treatment. Further validation with larger datasets is necessary to ensure robustness and generalizability.