Session Information
Date: Sunday, October 7, 2018
Session Title: Dystonia
Session Time: 1:45pm-3:15pm
Location: Hall 3FG
Objective: The objective of this study was to compare options for machine learning-based video analysis software for measuring motor severity in cervical dystonia (CD).
Background: Quantifying motor severity is critical for research into the pathophysiology of CD and as an outcome measure for clinical trials. However, measures such as the Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS) are based on human judgement, which is inherently subjective. Computational analyses of video recordings that capture head orientation offer the possibility of objectively measuring motor features of CD.
Methods: We analyzed digital videos recorded at 30 frames per second from CD patients demonstrating range of motion in each of three axes: pitch (antero/retrocollis), roll (laterocollis, or “tilt”), and yaw (torticollis). Videos were decomposed into individual frames, then head orientation estimated on each frame, quantified as degrees of rotation in each axis. We screened publicly-available software for the ability to measure head orientation and facial expressions from videos. Each option was analyzed for the percent of frames when the face was found, sensitivity to head rotations in each axis, the ability to find the face during a sensory trick involving touching the side of the face, and the ability to detect tremor.
Results: We identified and tested three software packages that met our criteria: Emotient’s FACET (previously known as CERT), Affectiva’s Affdex, and OpenFace. FACET, Affdex, and OpenFace were able to find the face in 55, 83, and 84% of the frames, respectively. Affdex and OpenFace were much more robust than FACET for finding the face during head tilt. All three packages could find the face during the sensory trick. FACET and Affdex are better at detecting head tremor. Collectively, Affdex appears best suited for CD.
Conclusions: This study provides a preliminary demonstration that computational video analysis can capture several of the motor phenomena associated with CD. Pending broader validation, it could form the basis for objective assessment that is complementary to rating scales, thereby reducing measurement noise for assessing CD severity. We are extending the findings from this study with a cohort of 200 patients with isolated CD originally recruited from 10 sites in North America through the Dystonia Coalition’s project to validate the Comprehensive Cervical Dystonia Rating Scale.
To cite this abstract in AMA style:
D. Peterson, Z. Zhang, J. Perlmutter, G. Stebbins, C. Comella. Toward an automated, video-based method to measure motor severity in cervical dystonia [abstract]. Mov Disord. 2018; 33 (suppl 2). https://www.mdsabstracts.org/abstract/toward-an-automated-video-based-method-to-measure-motor-severity-in-cervical-dystonia/. Accessed October 6, 2024.« Back to 2018 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/toward-an-automated-video-based-method-to-measure-motor-severity-in-cervical-dystonia/