As technology advances, it has been assumed for a long time that verbal and physical diagnoses for mental health will eventually be replaced by computers and brain scans.
However, there is currently no blood or genetic test that can accurately diagnose mental illnesses, especially between illnesses that have overlapping symptoms. Nevertheless, finding an objective method to accurately diagnose mental illnesses would save a lot of time (for the patients and psychiatrists and allow earlier treatment), as well as money (when the wrong treatments or medications are prescribed). Currently, diagnostic models in psychiatry can be faulty for three reasons:
- Patient heterogeneity: the patient’s psychological state, their ability to provide reliable information and individual differences in clinical presentation.
- Clinical inconsistency: psychiatrists can have different opinions on the same patients.
- Nomenclature inadequacy: psychiatric illnesses can be classified differently depending on which models or classifications are used by the psychiatrist.
A recent study from the University of Tokyo has demonstrated the likelihood of this technological replacement when looking at individuals with schizophrenia or autism spectrum disorder (ASD). These two disorders were selected as ASD patients have a 10x higher risk of developing schizophrenia than the general population. However, the treatments for these conditions differ, therefore accurate diagnoses are essential.
The researchers combined machine learning with brain imaging tools to create a method to accurately diagnose mental illnesses. Machine learning refers to the application of artificial intelligence (AI) that provides systems with the ability to automatically learn and improve from experience, without being explicitly programmed. Previous studies have successfully created machine learning algorithms to distinguish between patients and healthy controls; however, this is the first study that has successfully distinguished between psychiatric diagnoses.
In order to create the algorithm, the researchers trained the computer using (MRI) brain scans of 206 adults: individuals with ASD (36), schizophrenia (64) or typically developing individuals (106). The researchers made sure to use scans from the same MRI machine over the three years of data collection, as these can otherwise differ. The individuals ranged on both the schizophrenic and ASD spectrums. The computer was trained to find patterns in thickness, surface area and volume of areas in the brain (using 6 different algorithms for additional accuracy).
To finally test the algorithm, 43 new MRI scans were presented to the computer. The computer could distinguish between patients with schizophrenia, schizophrenia risk factors, ASD and non-patients with up to 85% accuracy!
The algorithm’s ability to detect schizophrenia risk factors suggests that brain structures change before symptoms even arise – this finding could be extremely useful to develop prevention treatments and for early care.
The researchers also found that the core indicators used by the algorithm were the thickness of the cerebral cortex, as well as subcortical volumes, suggesting that these are areas which future neuropsychiatric investigations should focus on.
Source: Yassin, et al. (August 2020). “Machine-learning classification using neuroimaging data in schizophrenia, autism, ultra-high risk and first-episode psychosis”. Nature: Translational Psychiatry. 10, 278.