Toward neuroimaging-based diagnostic support: a deep learning approach with a closed-loop system for psychiatric classification

While neuroimaging offers valuable insights into brain alterations associated with psychiatric disorders, its translation into routine clinical practice remains challenging. we developed the patch-based hierarchical network (PHN), a deep learning framework for classifying multiple psychiatric disorders from structural MRI. Trained on a large, curated dataset (n = 2,490) of four major disorders and controls, the PHN's generalizability was confirmed on independent research datasets (n = 1,346) and real-world clinical data (n = 344). The model demonstrated robust performance, showing promise in real-world evaluations by reflecting complex presentations such as comorbidities. A key contribution is the implementation and deployment of a closed-loop, neuroimaging-based diagnostic support system integrating the PHN into an active clinical workflow. This work provides a tangible step toward bridging the research-to-practice gap, leveraging artificial intelligence to provide objective support for psychiatric diagnosis.