Recently, functional magnetic resonance imaging (fMRI) has been employed to classify brain disorders such as Alzheimer’s disease and autism spectrum disorder. Deep learning models have made significant progress in interpreting complex neural data in the context of the evolving field of fMRI analysis. In this study, we face the substantial challenge of simultaneously processing connections between brain regions and contextual representations on different time scales. This study proposes STCTb, a spatio-temporal collaborative Transformer block structure for fMRI time series. STCTb uniquely integrates multi-scale spatiotemporal information processing and enables the collaboration of temporal and spatial features through an innovative architecture, aiming to improve the sensitivity and specificity of the model in recognizing brain activity signals. While inheriting the cascade structure of the Swin Transformer, the mechanism proposes a two-branch block structure, which achieves rich and efficient global information integration at both temporal and spatial scales. Extensive experiments on the publicly available fMRI datasets ADNI and ABIDE-I have shown that the model using STCTb exhibits significantly superior performance compared to existing methods, and this innovative approach provides valuable insights for the further development of deep learning in the field of functional data analysis of brain diseases.