Amyotrophic Lateral Sclerosis and Post-Stroke Orofacial Impairment Video-based Multi-class Classification

Abstract

Neurological diseases, such as ALS and Stroke, that affect the brain including the nerves found throughout the body including the spinal cord generally require various forms of testing and clinical diagnosis in order to detect. These current forms of diagnosis, however, present a limitation in the form of being either expensive or subjective. Research has been done in the area of automated medical assessment via machine learning with the goal of offering cheaper and more objective alternatives for aiding diagnosis. For the case of ALS and orofacial impairment in stroke, it has been shown that using features derived from facial movement in videos, it is possible to detect the presence of these neurological diseases among healthy patients, separately. Research in this area, however, is still relatively few and allows for exploration of improvements in the overall model, especially with the emergence of newer algorithms for detecting facial landmarks. For this research, the improvements to be explored in the model will come in the form of exploring how the model can be trained to detect both (multi-class) ALS and orofacial impairment in post-stroke among a healthy population. Results show that features calculated from facial landmarks in videos, it is possible to develop a single muti-class detection model ALS, and orofacial impairment in stroke among a healthy population with accuracy as high as 86%.