Cutting through the hype: artificial intelligence for clinical decision support in psychiatry

Cutting through the hype: artificial intelligence for clinical decision support in psychiatry

Mental health conditions are one of the most significant challenges to global society. This thesis explores the integration of artificial intelligence (AI) techniques in psychiatric research, focusing on enhancing the detection, diagnosis, and treatment of depression. While simultaneously exploring the implications of integrating AI into clinical practice. Through the use of empirical experiments the works contained within this thesis demonstrate the potential uses for AI in mental healthcare. These works show AI techniques can reliably predict treatment outcomes to repetitive transcranial magnetic stimulation (rTMS) above the existing state-of-the-art. Combining these methods with explainable AI (XAI) this work identifies candidate biomarkers indicative of treatment response to rTMS. Furthermore, this work shows predictive performance can be improved by using diversity enhanced training data. This thesis includes a novel method for enhancing the diversity of training data. Including experiments which demonstrate the possibility of synthetic data to improve dataset diversity. This thesis also presents a novel method for addressing label bias when detecting suicide risk on social media through semi-supervised deep label smoothing. Empirical experiments show this methods improves classification accuracy by leveraging fuzzy labels and Bayesian techniques. Put together, the research within this thesis highlights the transformative potential of AI in psychiatry, demonstrating the possibility of personalised psychiatry, advocating for innovative data augmentation and regularisation methods to improve model performance. By critically analysing these empirical experiments, this thesis examines the broader implications of AI in psychiatry. It places special emphasis on methods to ensure the ethical and equitable deployment of AI in mental healthcare.

File reproduced in accordance with the copyright policy of the publisher/author/creator.