Video-based Parkinson's disease detection in low data regimes

Sriram, Pranav

Permalink

https://hdl.handle.net/2142/121283 Copy

Description

Title

Video-based Parkinson's disease detection in low data regimes

Author(s)

Sriram, Pranav

Issue Date

2023-07-21

Director of Research (if dissertation) or Advisor (if thesis)

Wang, Yuxiong

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Computer vision
• Deep learning

Abstract

Parkinson’s disease is a prevalent neurodegenerative disorder affecting mil- lions of people worldwide. Its characteristic symptoms manifest themselves in an increasingly severe fashion as the disease progresses. Thus, early diag- nosis is vital in slowing the disease progression and initiating the appropriate treatment in a timely manner. Moreover, Parkinson’s disease detection is a nontrivial task, encompassing many multimodal cues such as speech, appear- ance and muscle movement for accurate diagnosis. The difficulty of the task opens up the avenue of AI-enhanced disease detection through intelligent, state-of-the-art models leveraging all the aforementioned cues. Existing works have explored a variety of different techniques for Parkin- son’s disease diagnosis, ranging from monitoring breathing signals to tracking muscular movement for tremor and hypomimia detection. However, these works all use their own proprietary medical datasets, which are not released due to patient confidentiality. As a result, the performance of models across different works cannot be compared, hindering the progress towards strong performance on this task. Moreover, existing techniques also do not leverage state-of-the-art vision or deep learning methodologies and rather use simpler deep learning models with reduced learning capabilities. For our methodology development, we construct our novel dataset PD- Dataset composed of clips extracted from videos. These clips contain both healthy controls (celebrities without Parkinson’s) and Parkinson’s patients, providing a comprehensive benchmark for model evaluation. In addition, we propose a new detection model based on state-of-the-art video understanding architectures and demonstrate strong performance on our novel benchmark.

Graduation Semester

2023-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/121283

Copyright and License Information

Copyright 2023 Pranav Sriram

Owning Collections