OCD Diagnosis via Smoothing Sparse Network and Stacked Sparse Auto-Encoder Learning

Abstract

Obsessive-compulsive disorder (OCD) is a serious mental illness that affects the overall quality of patients’ daily life. Since sparse learning can remove redundant information in resting-state functional magnetic resonance imaging (rs-fMRI) data via the brain functional connectivity network (BFCN) and retain good biological characteristics, it is an important method for OCD analysis. However, most existing methods ignore the relationship among subjects. To solve this problem, we propose a smoothing sparse network (SSN) to construct BFCN. Specifically, we add a smoothing term in the model to constrain the relationship and increase the similarity among the subjects. As a kind of deep learning method, the stacked sparse auto-encoder (SSAE) can learn the high level internal features from data and reduce its dimension. For this reason, we design an improved SSAE to learn the high level features of BFCN and reduce the data dimension. We add a $\ell_{2}$-norm to prevent overfitting as well. We apply this framework on OCD dataset self-collected from local hospitals. The experimental results show that our method can achieve quite promising performance and outperform the state-of-the-art methods.

This work was supported partly by National Natural Science Foundation of China (Nos. 31871113, 61871274, 61801305 and 81571758), National Natural Science Foundation of Guangdong Province (No. 2017A030313377), Guangdong Pearl River Talents Plan (2016ZT06S220), Shenzhen Peacock Plan (Nos. KQTD2016053112051497 and KQTD2015033016 104926), and Shenzhen Key Basic Research Project (Nos. JCYJ2017 0413152804728, JCYJ20180507184647636, JCYJ20170818142347251 and JCYJ20170818094109846).