Mental health diagnoses and the role of machine learning

Machine learning could aid mental health diagnoses
© iStock/metamorworks

Researchers at the University of Birmingham have developed a way of using machine learning to aid mental health diagnoses, and more specifically, to more accurately identify patients with a mix of psychotic and depressive symptoms.

It is common for patients with psychosis or depression to experience symptoms of both conditions which has meant that traditionally, mental health diagnoses have been given for a ‘primary’ illness with secondary symptoms of the other.

Making an accurate diagnosis often poses difficulties to mental health clinicians and diagnoses often do not accurately reflect the complexity of individual experience or neurobiology. For example, a patient being diagnosed with psychosis will often have depression regarded as a secondary condition, with more focus on the psychosis symptoms, such as hallucinations or delusions; this has implications on treatment decisions for patients.

A team at the University of Birmingham’s Institute for Mental Health and Centre for Human Brain Health, along with researchers at the European Union-funded PRONIA consortium, explored the possibility of using machine learning to create extremely accurate models of ‘pure’ forms of both illnesses and using these models to investigate the diagnostic accuracy of a cohort of patients with mixed symptoms. The results of this study have been published in Schizophrenia Bulletin.

Paris Alexandros Lalousis, lead author, explains that “the majority of patients have co-morbidities, so people with psychosis also have depressive symptoms and vice versa… That presents a big challenge for clinicians in terms of diagnosing and then delivering treatments that are designed for patients without co-morbidity. It’s not that patients are misdiagnosed, but the current diagnostic categories we have do not accurately reflect the clinical and neurobiological reality”.

The researchers analysed questionnaire responses and detailed clinical interviews, as well as data from structural magnetic resonance imaging from a cohort of 300 patients taking part in the study. From this group of patients, they identified small subgroups of patients, who could be classified as suffering either from psychosis without any symptoms of depression, or from depression without any psychotic symptoms.

With the goal of developing a precise disease profile for each patient and testing it against their diagnosis to see how accurate it was, the research team was able to identify machine learning models of ‘pure’ depression, and ‘pure’ psychosis by using the collected data. They were then able to use machine learning methods to apply these models to patients with symptoms of both illnesses.

The team discovered that patients with depression as a primary illness were more likely to have accurate mental health diagnoses, whereas patients with psychosis with depression had symptoms which most frequently leaned towards the depression dimension. This may suggest that depression plays a greater part in the illness than had previously been thought.

Lalousis added: “There is a pressing need for better treatments for psychosis and depression, conditions which constitute a major mental health challenge worldwide. Our study highlights the need for clinicians to understand better the complex neurobiology of these conditions, and the role of ‘co-morbid’ symptoms; in particular considering carefully the role that depression is playing in the illness.

“In this study we have shown how using sophisticated machine learning algorithms, which take into account clinical, neurocognitive, and neurobiological factors can aid our understanding of the complexity of mental illness. In the future, we think machine learning could become a critical tool for accurate diagnosis. We have a real opportunity to develop data-driven diagnostic methods – this is an area in which mental health is keeping pace with physical health and it’s really important that we keep up that momentum.”

Subscribe to our newsletter


Please enter your comment!
Please enter your name here