Automatic detection of orofacial impairment in stroke

Stroke is a devastating condition that affects the ability of people to communicate through speech, leading to social isolation and poor quality of life. The quantitative evaluation of speech and orofacial movements is essential for assessing the impairment and identifying treatment targets. However, to our knowledge, a tool for the automatic orofacial assessment, which considers multiple aspects of orofacial impairment (e.g., range of motion in addition to asymmetry), has not been developed for this clinical population. In this work, we tested a video-based approach for the automatic orofacial assessment in stroke survivors, combining low-cost depth sensor and face alignment algorithms for extracting facial features. Twelve patients post-stroke and 11 control subjects were evaluated during speech and non-speech tasks. By using a small feature-set representing range of motion and asymmetry of face movements, it was possible to discriminate patients post-stroke from control subjects with high accuracy (87%). Further insights on the choice of the task and face alignment algorithm are provided, demonstrating that a non-parametric approach such as SDM can provide better results. Through this work we demonstrated the feasibility of an objective tool to support clinicians in the assessment of speech and orofacial impairment post-stroke.