Deep Learning-based Pose Estimation for Dystonia Score Prediction

Submitted by Sushant Gautam to Department of Electronics and Computer Engineering, Thapathali Campus

(Institute Of Engineering, Tribhuvan University)

on Partial Fulfillment of the requirement for the degree of Master of Science In Informatics And Intelligent Systems Engineering

Supervised By:

Dr. Bishesh Khanal https://www.naamii.org.np/teams/bishesh-khanal/ Dr. Nabin Koirala https://haskinslabs.org/people/nabin-koirala Dr. Ajad Chhatkuli https://ee.ethz.ch/the-department/people-a-z/person-detail.MjM3Njk0.TGlzdC8zMjc5LC0xNjUwNTg5ODIw.html

Report and Presentations:

Report: https://raw.githubusercontent.com/SushantGautam/DystViz/master/defense-files/1.%20Defense_Report_Sushant_Deep%20Learning%20based%20Pose%20Estimation%20for%20Dystonia%20Score.pdf

Presentaion: https://github.com/SushantGautam/DystViz/blob/master/defense-files/2.%20Defense_Slide_Sushant_Deep%20Learning%20based%20Pose%20Estimation%20for%20Dystonia%20Score.pdf

Outline:

Videos of patients from Dystonia Colatition is processed with OpenPose

The temporal displacement of the pose coordinates are plotted:

The temporal spatial displacement is fed to CNN along with the clinical score:

Difference in Model Score and Clinical Score showing that most of the scored are accurate as almost all difference is around zero.