Evolutionary neural architecture search for automated MDD diagnosis using multimodal MRI imaging

Tongtong Li; Ning Hou; Jiandong Yu; Ziyang Zhao; Qi Sun; Miao Chen; Zhijun Yao; Sujie Ma; Jiansong Zhou; Bin Hu

doi:10.1016/j.isci.2024.111020

Evolutionary neural architecture search for automated MDD diagnosis using multimodal MRI imaging

iScience. 2024 Sep 24;27(10):111020. doi: 10.1016/j.isci.2024.111020. eCollection 2024 Oct 18.

Authors

Tongtong Li^{1

2}, Ning Hou³, Jiandong Yu^{1

2}, Ziyang Zhao^{1

2}, Qi Sun^{1

2}, Miao Chen^{1

2}, Zhijun Yao^{1

2}, Sujie Ma⁴, Jiansong Zhou⁵, Bin Hu^{1

2

6

7

8}

Affiliations

¹ School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China.
² Gansu Provincial Key Laboratory of Wearable Computing, Lanzhou University, Lanzhou 730000, China.
³ Medical Department, The Third People's Hospital of Tianshui, Tianshui 741000, China.
⁴ Sleep Department, The Third People's Hospital of Tianshui, Tianshui 741000, China.
⁵ National Clinical Research Center for Mental Disorders, and National Center for Mental Disorders, The Second Xiangya Hospital of Central South University, Changsha 410000, China.
⁶ School of Medical Technology, Beijing Institute of Technology, Beijing 100081, China.
⁷ CAS Center for Excellence in Brain Science and Intelligence Technology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai 200031, China.
⁸ Joint Research Center for Cognitive Neurosensor Technology of Lanzhou University & Institute of Semiconductors, Chinese Academy of Sciences, Lanzhou 730000, China.

Abstract

Major depressive disorder (MDD) is a prevalent mental disorder with serious impacts on life and health. Neuroimaging offers valuable diagnostic insights. However, traditional computer-aided diagnosis methods are limited by reliance on researchers' experience. To address this, we proposed an evolutionary neural architecture search (M-ENAS) framework for automatically diagnosing MDD using multi-modal magnetic resonance imaging (MRI). M-ENAS determines the optimal weight and network architecture through a two-stage search method. Specifically, we designed a one-shot network architecture search (NAS) strategy to train supernet weights and a self-defined evolutionary search to optimize the network structure. Finally, M-ENAS was evaluated on two datasets, demonstrating that M-ENAS outperforms existing hand-designed methods. Additionally, our findings reveal that brain regions within the somatomotor network play important roles in the diagnosis of MDD, providing additional insight into the biological mechanisms underlying the disorder.

Keywords: Bioinformatics; Neuroscience; Psychology.