Investigating the Heterogeneity of Depression and Psychosis and Predicting Recovery Using Neuroimaging and Metabolic Biomarkers: a Multimodal Machine Learning Approach
Depressive and psychotic disorders constitute the major challenges of mental ill health to the world’s population. Depression is the leading cause of disability world-wide with 300 million people being affected. Adolescents and young people are particularly vulnerable to developing this illness. In the UK, the one-year prevalence of depression in adolescents is 5%. Schizophrenia is one of the most common psychotic disorders. It contributes to 13.4 million years lived with disability globally and costs around US $94-102 billion in the USA and €94 billion in Europe each year.
Both disorders present clinicians with challenges in terms of treatment. The majority of patients experience only partial remission and have recurrent relapses. Furthermore, both disorders share common clinical, neurobiological, and neurocognitive characteristics. Complex psychopathology presents clinicians with the challenge of correctly identifying co-morbidities, avoiding misdiagnoses, and tailoring therapeutic options for the individual. Currently, diagnosis is based on the phenomenological evaluation of symptoms and behaviour. However, there remains significant debate about the heterogeneity of these disorders. Neuroimaging and metabolic markers hold significant diagnostic and prognostic potential to aid these difficulties. Using machine learning can help disentangle complex psychopathology and provide a deeper understanding of the aetiology of depression and psychosis.
Data from the PRONIA study, an EUFP7 funded 8 centre study (https://www.pronia.eu/) recruiting recent onset depression and recent onset psychosis participants will be used to identify common and distinct neuroimaging, clinical, and metabolic features of depression and psychosis across diagnostic groups. Sophisticated machine learning approaches will be used to elucidate the heterogeneity of both disorders. Features that provide significant information will be used to build predictive machine learning models that predict recovery from depression and psychosis as well as diagnostic machine learning models.
Supervisors:
Professor Rachel Upthegrove
Professor Stephen Wood
Professor Nikolaos Koutsouleris
Dr. Lianne Schmaal
Dr. Renate Reniers