RobIn: A robust interpretable deep network for schizophrenia diagnosis

Schizophrenia is a severe mental health condition that requires long and complicated diagnostic process. However, early diagnosis vital to control symptoms. Deep learning has recently become popular way analyse interpret medical data. Past attempts use deep for schizophrenia from brain-imaging data have shown promise but suffer large training-application gap — it difficult apply lab research the real world. We propose reduce this by focusing on readily accessible collect set of psychiatric observations patients based DSM-5 criteria. Because similar already recorded in all clinics diagnose using DSM-5, our method could be easily integrated into current processes as tool assist clinicians, whilst abiding formal To facilitate real-world usage system, we show interpretable robust. Understanding how machine reaches its essential allow clinicians trust diagnosis. framework, fuse two complementary attention mechanisms, ‘squeeze excitation’ ‘self-attention’, determine global attribute importance interactivity, respectively. The model uses these scores make decisions. This allows understand was reached, improving model. models often struggle generalise different sources, perform experiments with augmented test evaluate model’s applicability find more robust perturbations, should therefore better clinical setting. It achieves 98% accuracy 10-fold cross-validation.