Machine Learning Based Approach for Neurological Disorders in Resting-State Mice and Humans using Multi-Channel NIRS System
다채널 근적외선분광법 시스템 및 머신러닝을 이용한 휴지 상태 쥐와 사람의 신경계 장애 연구
- 주제(키워드) Machine Learning , NIRS , Parkinson's disease , Depression
- 발행기관 고려대학교 대학원
- 지도교수 성준경
- 발행년도 2020
- 학위수여년월 2020. 8
- 학위구분 석사
- 학과 대학원 바이오융합공학과
- 원문페이지 47 p
- UCI I804:11009-000000231931
- DOI 10.23186/korea.000000231931.11009.0001166
- 본문언어 영어
- 제출원본 000046048275
초록/요약
Neurological disorders such as Parkinson’s disease (PD) and depression can impair neurovascular structure and change cerebral hemodynamics. Near-infrared spectroscopy (NIRS) can resolve hemodynamic changes, including oxy-hemoglobin, deoxy-hemoglobin and total hemoglobin changes. In this work, I explore the possibility of automatically detecting diagnostic markers using machine learning for patients with PD and mice with depression, specifically in a taskless state. Finding such markers is critical because such patients often have difficulty in performing even simple tasks correctly due to their neurological condition. First, I test the feasibility of classifying resting-state NIRS signals of depressive-like behavior mouse and control mouse, after performed simple stimulation experiments to verify the NIRS animal system. The result indicates that classification accuracy can reach up to 96.78±0.73% and that hemodynamic changes of the anterior part of the brain and cerebral asymmetry due to depression both can have a large effect on classification accuracy. Then, I applied machine learning algorithms for PD patients undergoing tests of autonomic dysfunction. I tested combinations of NIRS input data to increase the classification accuracy between PD patients with orthostatic hypertension (OH), PD patients without OH, and patients who did not have either PD or OH. The result indicates that accuracy over 85% can be achieved by optimizing time-series input. Furthermore, the use of a channel correlation matrix as input can increase the accuracy to over 90%. These studies showcase the abilities of NIRS-based inputs paired with machine learning algorithms to accurately classify neurological disorders such as PD and depression.
more목차
1. Introduction
1.1 Motivation
1.2 Near-infrared spectroscopy
1.3 Depression
1.4 Parkinson’s disease and orthostatic intolerance
1.5 Machine learning approach for NIRS
2. Experiment and method
2.1 Depression in mice
2.1.1 NIRS system and verification
2.1.2 Experiment protocol
2.2 Parkinson’s disease in human
2.2.1 NIRS system
2.2.2 Experiment protocol
2.3 Input and dataset
2.4 Machine learning and cross-validation methods
3. Experiment result
3.1 Depression in mice
3.2 Parkinson’s disease in human
4. Discussion and conclusion
References
