Motion Analysis for Clinical Classification of Dystonia Patients using deep Learning-based Approach

Motion Analysis for Clinical Classification of Dystonia Patients using deep Learning-based Approach

Motivation: We aim to design and implement a deep learning-based algorithm to classify different dystonia types and predict their severity by detecting patients’ gaits and poses using single-camera videos. This study’s overarching goal would be to develop an objective diagnostic tool that could be used in clinical settings. 

Research Questions:

  1. Could a deep learning-based model be able to classify different forms of dystonia?
  2. Can we predict the dystonia severity scores using this model?
  3. Would this model perform better or worse than the clinician’s diagnosis?

Brief Description: Dystonia is a neurodegenerative disorder characterized by involuntary muscle contractions that cause slow, repetitive movements, abnormal postures, tremors, and several other non-motor symptoms. There are several different forms of dystonia affecting one or groups of muscles throughout the body. The cause of Dystonia is still unknown. More importantly, the diagnosis of focal dystonia is complex and based on clinical settings, which is often prone to subjective bias. Hence, in this study, we would explore the possibility of using deep learning-based computer vision techniques to obtain clinically relevant features that could be used for unbiased classification of different forms of Dystonia and predict its severity. We will explore gait analysis, automatic pose, motion estimation, and several latest algorithms to develop a predictive model.

Research Themes: Transforming Global Health with Affordable & Inclusive AI
Project Category: Computer Vision