Summary
International Symposium on Nonlinear Theory and its Applications
2017
Session Number:C2L-D
Session:
Number:C2L-D-6
Deep Neural Generative Model for fMRI Image Based Diagnosis of Mental Disorder
Tetsuo Tashiro, Takashi Matsubara, Kuniaki Uehara,
pp.700-703
Publication Date:2017/12/4
Online ISSN:2188-5079
DOI:10.34385/proc.29.C2L-D-6
PDF download (497.4KB)
Summary:
Diagnosis of mental disorders based on fMRI brain image analysis often has two steps: unsupervised feature extraction and supervised classification. This is expected to prevent overfitting due to a small size of medical fMRI. However, the unsupervised feature extraction has a risk of extracting individual variability (such as brain shape) as a feature instead of disease-related brain activity. In this study, we propose an fMRI brain image analysis method based on conditional variational auto-encoder (CVAE), which is a deep learning model extracting features with given label information. The CVAE can classify fMRI images without another feature extraction process, suppresses overfitting, and achieves better diagnosis accuracy.