Leveraging Deep Learning For Automated Detection Of Mental Disorders: A Survey And Future Directions

Abstract

The worldwide burden of mental disorders such as bipolar disorder, schizophrenia, depression, and anxiety now impacts millions of people, making them a major global health issue. The diagnosis of mental illnesses is a critical determinant of effective treatment and therapy strategies require initiation before accurate identification can be made. Artificial intelligence and in particular, deep learning have recently been adopted to carry out the automation of mental disorder identification from multimodal data including text, speech, and neuroimaging. This survey presents a comprehensive review of current approaches in deep learning techniques for mental health analysis and mitigation, including transformer models, recurrent neural networks (RNNs), convolutional neural networks (CNNs) and hybrid architectures. We review their application in different data sources including audio, medical documents, online social network posts, and brain signals such as EEG and MRI. We also outline some major challenges including data scarcity, ethical concerns, model interpretability issues and generalization challenges. In addition, we outline future research directions which include multimodal fusion, explainable AI, privacy preserving federated learning and real time mental health monitoring. The aim of this work is to serve as a starting point for scholars and practitioners on the application of deep learning to improve the diagnosis of mental health. This paper aims to present an overview of the deep learning methods for detection of mental disorders, their application, challenges and the possible future directions.