The World Health Organisation has previously announced a significant piece of research outcome: 13% of the global burden of disease comprises of mental health problems and 25% of the population experiences a minimum of one mental health problem episode during their course. What is even more significant is that, to date, there are no identified bio-markers that can be used for the diagnosis of mental health disorders; and treatment plans don’t work successfully for the majority of patients, and/or cause severe adverse effects.
In the next 5 years, we propose to address the problem of mental health disorder analysis, and aid the implementation of a precision medicine framework which offers personalised treatment choices to increase the success rate of treatment. For that, we plan to make simultaneous use of multiple sources of information: ii) neuroimaging data ii) clinical/behavioural data (e.g. socio-demographics, lifestyle, family history, physical measures, cognitive function, health outcomes).
The first step of our approach would involve the discovery of associations between these different sources of information. For example, we may uncover some specific regions of the brain are highly associated with the joint factors of socio-demographics and family history of mental health diseases. After identifying these patterns, the second step would involve making use of them for the identification of patients at risk. For example, subjects whose data don’t follow the extracted relationship between the brain and socio-demographics & family history, can be flagged as carrying a risk of developing a disorder.
As we expect to extract multiple brain-behaviour associations/relations, patients can be clustered into different categories depending on which associations they are identified as “at risk” at. In other words, we aim to determine different types of risk factors for patients according to multiple aspects of mental health, so as to provide guidelines for specialised treatment.
While implementing our approach, in addition to the established traditional machine learning methods, we aim to utilise state-of-the-art deep learning (DL) algorithms. The comprehensive data collected from an extensive number of participants in the UK Biobank data set will therefore provide a means to accomplish our goal of implementing intelligent, novel and innovative DL approaches, which is not possible using smaller sized data sets in the field.
The practical outcomes of our inter-disciplinary research is anticipated to highly benefit the neuroscience and psychiatry communities, while the theoretical novelties would contribute to the computer science and machine learning fields.