Prediction of expanded disability status scale subscores of motor dysfunction in multiple sclerosis using depth-sensing computer vision

  • M. D'Souza ,
  • J. Burggraaff ,
  • P. Kontschieder ,
  • J. Dorn ,
  • C.P. Kamm ,
  • S. Seinheimer ,
  • P. Tewarie ,
  • C. Morrison ,
  • ,
  • ,
  • F. Dahlke ,
  • B Uitdehaag ,
  • L. Kappos

Multiple Sclerosis Journal | , Vol 21: pp. 409-409

Background: Clinical assessment of impairment and disability in Multiple Sclerosis (MS) remains the most important outcome in therapeutic trials, and is commonly assessed with the Expanded Disability Status Scale (EDSS). However, the EDSS exhibits high inter- and intra-rater variability. The ASSESS MS system is being developed as a non-invasive, more consistent and potentially finer grained tool to measure motor dysfunction in MS, by combining recordings of prescribed neurological movements with machine learning methods to assess motor dysfunction based on EDSS subscores.

Objectives: To test the prediction of EDSS subscores from recordings of a depth-sensing video analysed by machine learning algorithms.

Methods: Pre-defined movements from the EDSS assessment were recorded in 300 patients and 200 healthy volunteers. Video recordings of patients were scored by four neurologist from 3 sites based on the Neurostatus/EDSS assessment definitions. These scores were used to train a machine learning algorithm to correctly predict motor dysfunction from depth-sensing video recordings, which capture a 3D view of the patient.

Results: We report that on movements covering upper extremities and trunk, the machine learning algorithm predicts motor dysfunction of patients with MS with an accuracy similar to neurologists’ intra-rater retest reliability. For example, the agreement of the upper extremity tremor/dysmetria subscore from the finger-to-nose test of the algorithm with the neurologists’ assessment is 73% across scores 0, 1, 2 and 3, while the long-term intra-rater agreement of the neurologists with their previous assessment is 67% in this challenging-to-assess range.

Conclusions: Automated quantification of movement recordings using a depth-sensing camera and image analysis based on a machine-learning algorithm enables an accurate and sensitive quantitative assessment of motor dysfunction in MS patients. ASSESS MS is expected to improve the evaluation of disability progression in clinical studies and clinical practice.