The recent rapid progress in artificial intelligence (AI) and neuroimaging has paved the ways for exploring brain function and disorders. Nevertheless, the vast majority of current brain mapping methods are confined to specific brain regions, small sample sizes, high computation costs, and low clinical interpretability. In this work, we present a systematic mechanism of brain mapping, utilizing functional magnetic resonance imaging (fMRI) combined with AI-analysed methods, which may provide alternative approaches to address the above challenges. The model was designed to integrate multimodal neuroimaging data, which comprises DTI, EEG and MEG, with fMRI, offering a complete picture of neuronal activity. Our architecture employs state-of-the-art transformer-based deep learning models to strike the right balance between computational efficiency and prediction accuracy which is also interpretable at a clinical setting using explainable AI such as attention and SHAP. The framework is applicable to dynamic longitudinal modeling which enables the system to regularly predict disease evolution and treatment responses. Target populations To ensure model robustness, and cross-population generalizability, we trained the model on diverse, large-scale training datasets. Moreover, the proposed framework eases the process of automatic multimodal data fusion and on-the-fly neuroimaging processing, so as to be clinically useful. Experimental results indicate better performance in multi-domain brain mapping, precise diagnosis and personalized neuromodulation planning when compared to the state of art approaches. This is a significant step forward in the development of scalable, interpretable, and clinically applicable AI-driven brain mapping tools.