Abstract
There has been increasing interest in performing psychiatric brain imaging studies using deep learning. However, most studies in this field disregard three-dimensional (3D) spatial information and targeted disease discrimination, without considering the genetic and clinical heterogeneity of psychiatric disorders. The purpose of this study was to investigate the efficacy of a 3D convolutional autoencoder (3D-CAE) for extracting features related to psychiatric disorders without diagnostic labels. The network was trained using a Kyoto University dataset including 82 patients with schizophrenia (SZ) and 90 healthy subjects (HS) and was evaluated using Center for Biomedical Research Excellence (COBRE) datasets, including 71 SZ patients and 71 HS. We created 16 3D-CAE models with different channels and convolutions to explore the effective range of hyperparameters for psychiatric brain imaging. The number of blocks containing two convolutional layers and one pooling layer was set, ranging from 1 block to 4 blocks. The number of channels in the extraction layer varied from 1, 4, 16, and 32 channels. The proposed 3D-CAEs were successfully reproduced into 3D structural magnetic resonance imaging (MRI) scans with sufficiently low errors. In addition, the features extracted using 3D-CAE retained the relation to clinical information. We explored the appropriate hyperparameter range of 3D-CAE, and it was suggested that a model with 3 blocks may be related to extracting features for predicting the dose of medication and symptom severity in schizophrenia.
Highlights
Deep learning (DL) has dramatically improved technology in speech recognition, image recognition, and many other fields (LeCun et al, 2015)
We have shown that (1) the proposed 3D convolutional autoencoder (3D-CAE) successfully reproduced 3D magnetic resonance imaging (MRI) data with sufficiently low errors, and (2) the diagnostic label-free features extracted using 3D-CAE retained the relation of various clinical information
We explored the appropriate hyperparameter range of 3D-CAE, and our results suggest that a model with 3 blocks-based features might preserve information related to the medication dose and the severity of positive symptoms in patients with schizophrenia
Summary
Deep learning (DL) has dramatically improved technology in speech recognition, image recognition, and many other fields (LeCun et al, 2015). As the global burden of psychiatric disorders increases (Olesen et al, 2012; Whiteford et al, 2013), psychiatric brain imaging studies using DL are anticipated to bring many benefits to society (Vieira et al, 2017). There are two major concerns about applying DL to psychiatric brain imaging: (1) treatment of the high dimensionality of data, and (2) the heterogeneity of psychiatric disorders (Feczko et al, 2019). Region of interest (ROIs), one of the most popular feature extraction methods, has contributed to detecting various structural and functional abnormalities in the brains of patients with psychiatric disorders (Fornito et al, 2012; Fusar-Poli et al, 2012; Linden, 2012; Ratnanather et al, 2013). A limitation of these studies is that they ignore the 3D spatial information contained within the original MRI scans
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
