Back to Research

Journal of Parkinson’s Disease · Accepted 31 August 2022 · DOI 10.3233/JPD-223493

An Evaluation of KELVIN, an Artificial Intelligence Platform, as an Objective Assessment of the MDS UPDRS Part III

Krista Sibley, Christine Girges, Joseph Candelario, Catherine Milabo, Maricel Salazar, John Onil Esperida, Yuriy Dushin, Patricia Limousin & Thomas Foltynie

Research Overview

Motor severity in Parkinson’s disease is usually graded with the MDS-UPDRS, a scale applied by trained human examiners. Because that judgement is observational, the same patient can be scored differently by two raters, or by the same rater on two occasions. This study asked how closely an automated video-analysis platform, KELVIN, tracks that clinical judgement.

Twenty-two people with Parkinson’s disease were assessed by two experienced MDS-UPDRS raters at the same time, and the video recorded during each of those two assessments was additionally scored by KELVIN. That design yields paired clinician scores and paired platform scores for the same patient sessions, measured both OFF and ON dopaminergic medication.

Across a seven-item subset of MDS-UPDRS part 3, agreement between the platform’s two scores was reported as an intraclass correlation coefficient of 0.80 OFF medication and 0.73 ON medication. The two clinicians agreed more closely with each other, at 0.99 and 0.94 respectively, which the authors note may reflect the experience of this particular rating team. Comparing the averaged clinician score with the averaged platform score for the same seven items gave 0.84 OFF medication but 0.31 ON medication; the authors suggest dyskinesia may have affected the platform’s scores in the ON state.

The authors conclude that KELVIN shows promise for capturing and scoring several MDS-UPDRS part 3 items at a level of consistency approaching that of experienced raters, and that the approach merits further investigation.

This summary was written by Machine Medicine Technologies and is not the published abstract. For the article itself, including its full methods, figures and tables, please read it at the publisher.

Original Publication

Originally published in Journal of Parkinson’s Disease (2022) as “An Evaluation of KELVIN, an Artificial Intelligence Platform, as an Objective Assessment of the MDS UPDRS Part III”. © 2022 IOS Press. All rights reserved.