Investigating Deep Learning for Classification of Psychiatric Disorders on Brain MRI Data Using GAN-Based Augmentation

One of the difficulties in diagnosing patients in the field of psychiatry is the lack of objective biomarker measurements. In general, diagnosis is based on traditional methods relying on behavioral symptoms. Currently, diagnostic approaches predominantly rely on traditional methods grounded in behavioral symptoms. The evolution of neuroimaging techniques has prompted extensive investigations into biological indicators of psychiatric disorders. Among these endeavors, the application of deep learning for classification has demonstrated substantial promise. Through generative models like the generative adversarial network (GAN), deep learning is being investigated more in the field of medical image analysis to address the problems of small data samples and class imbalance. In this study, deep learning is implemented for automatic detection of psychiatric diseases from brain magnetic resonance imaging (MRI). Convolutional neural networks (CNN) are used for the diagnosis to categorize brain images into one of the classes: (1) normal, (2) schizophrenia, and (3) bipolar disorder. Image synthesis as data augmentation was performed with GAN. The suggested approach aims to advance the state-of-the-art in psychiatric diagnosis and promote additional investigation into the use of brain imaging in diagnostic plans. The findings show that the amount of data and subject variability significantly affect the efficacy of approaches for diagnosing psychiatric disorders. In this study, the data scarcity problem CNN training was not successfully resolved by GAN-based augmentation. It is determined that in order to develop a successful automatic psychiatric disease diagnostic system, a sufficient data sample must be collected while taking into account variables like subject variability.