"Schizophrenia Multimodal Dataset and Code"

"Automated diagnosis of complex neurological disorders such as schizophrenia (SZ) from heterogeneous data remains a critical challenge in multimodal learning. Traditional approaches rely on subjective clinical evaluations and fail to exploit cross-modal correlations. We constructed a comprehensive multimodal dataset comprising resting-state functional MRI (rs-fMRI), three eye-tracking paradigms, and cognitive battery scores from 63 participants (32 SZ patients, 31 healthy controls) in an exploratory single-site validation study.Graph theory was applied to compute whole-brain and sub-network global efficiency (GE) and to analyze correlations with cognitive deficits. We propose an Improved Graph Attention Network (I-GAT) to encode brain connectivity and spectral features. Furthermore, a multimodal framework, DeepSeek-MMC, was developed using a pretrained large language model (LLM) with token projection modules to fuse heterogeneous features for classification."