State Space Model-based Classification of Major Depressive Disorder Across Multiple Imaging Sites

Shusheng Li, Bo Yang, Ting Ma, Chenfei Ye

Keywords: Classification, Major depressive disorder, Time series, State space model

Abstract:

Major Depressive Disorder (MDD) is a prevalent psychiatric condition characterized by persistent sadness and cognitive impairments, with high recurrence rates. This paper presents a novel state space model to classify MDD using BOLD time-series data, a brain function indicator represented as a vector of fMRI signals in the time domain. The analysis is based on data from 1642 subjects, encompassing
multiple imaging sites. We propose an innovative model that leverages Mamba, a state-space model, to capture long-term dependencies in multivariate time series data while maintaining linear scalability. The model exploits the unique properties of time series data to produce salient contextual cues at multiple scales, utilizing an integrated Mamba architecture to unify the handling of channel-mixing and channel-independence situations. This approach enables effective selection of contents for prediction against global and local contexts at different scales. The experimental results demonstrate that the proposed model achieves an average classification accuracy of 69.91%, significantly improving diagnostic accuracy for MDD compared to other approaches.

Submission Number: 32