Objective: To develop a video-based machine rating system for the Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) III.

Background: MDS-UPDRS III is an essential evaluation scale of motor symptoms for Parkinson’s disease (PD) patients. Manual rating of the scale is time-consuming and subjective and might be solved by machine rating. Previous studies have made remarkable achievements, but rating the entire spectrum of MDS-UPDRS III through video has never been achieved, which largely compromises its application in clinical use. Five ratings for “Rigidity” and one rating for “Postural Stability” of the MDS-UPDRS III requiring physical contact between the physician and the patient cannot be performed in machine vision. Our team previously managed to construct models for rigidity evaluation based on features collected from patients’ motions through machine vision, which makes machine rating of the entire MDS-UPDRS III possible [1].

Method: Features from 2,610 videos of 149 PD patients were collected for Extreme Gradient Boosting machine learning. Four indirect Subitem models, 18 direct Subitem models, 4 SubScale score models, and 1 Total score model were constructed [figure 1, figure 2].

Results: This machine rating system achieved a 95% accuracy rate in the majority (77.8%) of Subitems under clinically accepted standards. Eleven Subitem Rating Models (50%) achieved intraclass correlation coefficients (ICCs) greater than 0.75 (excellent) [table 1]. Seven Subitem Rating Models (32.8%) exhibited ICCs between 0.60 and 0.74 (good). The models for the SubScale scores (r (Pearson correlation coefficient) ≥0.83, mean absolute error (MAE)≤2.18, ICC>0.69) and the Total score (r=0.93, MAE=3.98, ICC=0.91) all performed well [table 2].

Conclusion: All models showed consistency with expert ratings. Video-based features were demonstrated to be more informative, and machine rating of MDS-UPDRS III is possible. Multicenter studies are needed for validation.

Figure 1

Figure 2

Table 1

Table 2

References: [1] Zhu X, Shi W, Ling Y, Luo N, Yin Q, Zhang Y, et al. Contactless evaluation of rigidity in Parkinson’s disease by machine vision and machine learning. Chin Med J 2023;136:2254–2256. doi: 10.1097/CM9.0000000000002668.

« Back to 2024 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/video-based-machine-rating-model-for-mds-updrs-iii-in-parkinsons-disease-a-proof-of-concept-pilot-study/