Pathological facial weakness detection using computational image analysis

Stroke is the leading cause of disability in the U.S. Early recognition and treatment of stroke decreases the mortality and possibility of severe injury. One prominent stroke symptom is facial weakness. Existing methods such as telestroke systems, which require complex configuration including a tablet, a dedicated video conference system, 4G wireless connection, and a standby clinical support team for diagnosis, are far from widely practical use. In this pilot study, we present an automatic pathological facial weakness detection tool based on computational image analysis. The proposed system is able to extract the facial landmarks and classify facial weakness using a learning method. In the experiment, each image in the dataset is scored independently by two senior neurology residents. Only images rated concordantly by both raters as likely normal or likely abnormal (including left facial weakness and right facial weakness) were included for analysis. Our method performed with overall accuracy of 94.5%, precision of 94.8%, sensitivity of 94.6%, and specificity of 96.8%. The experimental result indicates that our method is able to identify pathological facial weakness accurately on the static images.