TITLE:
Multi-Task Deep Learning for Predicting Depression Severity and Treatment Response Using Synthetic Neuroimaging: A Fully Controlled Simulation Framework
AUTHORS:
Rocco de Filippis, Abdullah Al Foysal
KEYWORDS:
Depression, Treatment Response Prediction, Multi-Task Deep Learning, Synthetic Neuroimaging, CNN, Brain Activation, Precision Psychiatry, Computational Psychiatry, Severity Estimation, Medical AI
JOURNAL NAME:
Open Access Library Journal,
Vol.12 No.8,
August
22,
2025
ABSTRACT: Major Depressive Disorder (MDD) remains a complex and debilitating psychiatric condition characterized by heterogeneous symptom profiles and substantial variability in treatment response. Despite extensive clinical research, the process of selecting effective pharmacological interventions for individual patients continues to rely heavily on trial-and-error, leading to prolonged suffering and increased healthcare costs. Neuroimaging-based predictive models have shown potential to support precision psychiatry; however, their progress is limited by the scarcity of large, annotated clinical datasets, high imaging costs, and strict patient privacy regulations. In this study, we propose a comprehensive synthetic neuroimaging and clinical data simulation framework that enables the development and testing of deep learning models in a fully controlled environment. Specifically, we designed a multi-task Convolutional Neural Network (CNN) that simultaneously predicts depression severity as a continuous outcome and treatment response as a binary classification task. The synthetic data were generated with biologically plausible brain activation patterns, targeting clinically relevant regions such as the prefrontal cortex, cingulate cortex, amygdala, and hippocampus. Clinical variables, including age, gender, baseline severity, and treatment group, were integrated into the simulation to reflect real-world heterogeneity. The proposed model achieved a Mean Absolute Error (MAE) of 2.84 for severity prediction and an Area Under the Curve (AUC) of 0.80 for treatment response classification on the held-out test set. These results demonstrate that synthetic-data-driven pipelines can effectively support the development of predictive models for complex psychiatric disorders. Our approach provides a scalable and reproducible pathway for accelerating algorithm validation, minimizing ethical constraints, and enabling large-scale experimentation in mental health research and computational psychiatry.