Systematic review finds facial expression recognition shows promise for stroke diagnosis and rehabilitation monitoring

BackgroundStroke remains a leading cause of long-term disability and mortality worldwide, the early identification and evaluation of rehabilitation effectiveness are of great significance. Facial expression recognition (FER), a computer vision technology, offers potential for early screening and monitoring in many conditions. However, its application in stroke remains underexplored.AimsTo offer insights into how the FER can be used in stroke identification and rehabilitation monitoring.MethodsWe systematically searched four databases, including PubMed, Web of Science, CNKI, and WanFang, for studies on FER applications in stroke, from database inception to January 2025.ResultsA total of 1,855 studies were identified, of which nine met the inclusion criteria (e.g., 8 diagnostic studies and 1 rehabilitation trial). Eight studies demonstrated FER’s diagnostic utility for stroke, achieving accuracies ranging from 82 to 98% through facial asymmetry analysis during standardized tasks in public/private datasets. Specific tasks such as KISS, SPREAD, and non-speech movements were particularly effective. One study achieved 99.81% accuracy in monitoring rehabilitation intensity by classifying real-time facial expressions to tailor training intensity. The data sources included images (66.7%) and clinical patients database (55.6%).ConclusionFER technology exhibits substantial potential as an auxiliary tool in stroke diagnosis and emerging rehabilitation applications by enabling precise analysis of mandibular and facial movements. Nevertheless, FER models face considerable challenges in real-world clinical translation. Future research could integrate multimodal data and in-the-wild databases to facilitate the clinical implementation of FER technology, thereby improving care delivery for both patients and clinicians and reducing patient mortality and disability.