Automatické hodnocení obličejových pohybů u pacientů trpících roztroušenou sklerózou

Abstract

Multiple Sclerosis (MS) is the most common non-traumatic, neurodegenerative disease affecting young adults worldwide with a total of 2.8 million people diagnosed. Main characteristics of the disease are the appearance of demyelinating plaque on neuronal structures. When demyelination occurs in the facial pathway, motor manifestations affecting the facial expression may arise which can include facial myokymia, hemifacial spasm or facial palsy. These all belong to the most common presentations of movement disorders in MS. Currently, there is no existing easy-to-use, objective and fully automatic tool enabling fast and reliable assessment of facial movement disruption for MS. The goal of this presented thesis is to develop a tool for the computerized, video-based assessment of facial disruption to facilitate an accurate, objective, easily applicable, and cost-effective method to evaluate facial movement in patients with MS. Forty native Czech-speakers have been recorded during a facial expressivity examination and one-minute-long video-recordings of their freely spoken monologue were used in the subsequent analysis. To analyze the disruption of facial movement, an end-to-end neural network based facial landmark detection algorithm “Face Mesh” was applied and a total of six facial movement markers have been proposed to parametrize asymmetric movement of the face. Significant differences were found in the face symmetry between MS patients with facial palsy and a healthy control group. Subsequently, a classification algorithm was trained using multinomial logistic regression that reached an AUC of 0.71. The results of this work confirmed the utility of an automated objective tool for facial disruptions in MS and present the disruption of facial movement as a possible disease biomarker. Moreover, the classification experiment emphasized the need of regional assessment in the evaluation of facial manifestations in MS.

Multiple Sclerosis (MS) is the most common non-traumatic, neurodegenerative disease affecting young adults worldwide with a total of 2.8 million people diagnosed. Main characteristics of the disease are the appearance of demyelinating plaque on neuronal structures. When demyelination occurs in the facial pathway, motor manifestations affecting the facial expression may arise which can include facial myokymia, hemifacial spasm or facial palsy. These all belong to the most common presentations of movement disorders in MS. Currently, there is no existing easy-to-use, objective and fully automatic tool enabling fast and reliable assessment of facial movement disruption for MS. The goal of this presented thesis is to develop a tool for the computerized, video-based assessment of facial disruption to facilitate an accurate, objective, easily applicable, and cost-effective method to evaluate facial movement in patients with MS. Forty native Czech-speakers have been recorded during a facial expressivity examination and one-minute-long video-recordings of their freely spoken monologue were used in the subsequent analysis. To analyze the disruption of facial movement, an end-to-end neural network based facial landmark detection algorithm “Face Mesh” was applied and a total of six facial movement markers have been proposed to parametrize asymmetric movement of the face. Significant differences were found in the face symmetry between MS patients with facial palsy and a healthy control group. Subsequently, a classification algorithm was trained using multinomial logistic regression that reached an AUC of 0.71. The results of this work confirmed the utility of an automated objective tool for facial disruptions in MS and present the disruption of facial movement as a possible disease biomarker. Moreover, the classification experiment emphasized the need of regional assessment in the evaluation of facial manifestations in MS.