Deep Generative Model of Individual Variability in fMRI Images of Psychiatric Patients

• Published in: IEEE transactions on biomedical engineering Vol. 68; no. 2; pp. 592 - 605
• Main Authors: Matsubara, Takashi, Kusano, Koki, Tashiro, Tetsuo, Ukai, Ken'ya, Uehara, Kuniaki
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.02.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Analytical models; Biological system modeling; Biomarkers; Brain modeling; Data models; deep generative model; Deep learning; Diagnosis; Functional magnetic resonance imaging; Gender; Hidden Markov models; Individual variability; Magnetic resonance imaging; Medical imaging; Mental disorders; Neuroimaging; psychiatric disorder; Variability
• ISSN: 0018-9294; 1558-2531; 1558-2531
• DOI: 10.1109/TBME.2020.3008707

Neuroimaging techniques, such as the resting-state functional magnetic resonance imaging (fMRI), have been investigated to find objective biomarkers of neuro-logical and psychiatric disorders. Objective biomarkers potentially provide a refined diagnosis and quantitative measurements of the effects of treatment. However, fMRI images are sensitive to individual variability, such as functional topography and personal attributes. Suppressing the irrelevant individual variability is crucial for finding objective biomarkers for multiple subjects. Herein, we propose a structured generative model based on deep learning (i.e., a deep generative model) that considers such individual variability. The proposed model builds a joint distribution of (preprocessed) fMRI images, state (with or without a disorder), and individual variability. It can thereby discriminate individual variability from the subject's state. Experimental results demonstrate that the proposed model can diagnose unknown subjects with greater accuracy than conventional approaches. Moreover, the diagnosis is fairer to gender and state, because the proposed model extracts subject attributes (age, gender, and scan site) in an unsupervised manner.