From subjective to objective
After the publication of the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), which is known to be problematic for the purposes of diagnosing depression due to unclear criteria and subjective measurements [1,2], the National Institute of Mental Health (NIMH) started their work on Research Domain Criteria (RDoC) , a new approach to studying mental disorders.
According to RDoC, mental disorders should be studied in the context of six different domains, such as cognitive systems and negative valence systems. As they state, “the goal is to understand the nature of mental health and illness in terms of varying degrees of dysfunction in general psychological/biological systems” . In our project, we implement an RDoC approach to study depression.
In order to determine the proper treatment for patients, it is necessary to have an instrument that will help to understand the neural and biological factors underlying the symptoms of depression. Providing this instrument is the goal of our company. However, we understand that we cannot simply use Drysdale et al.’s
approach, because fMRIs are not affordable for most patients and require a considerable amount of time to perform. Therefore, in our project, we plan to rely mostly on EEG, which can show the activity of all neural circuits involved in depression [6,7] and is a well-known, reliable, and affordable instrument for clinicians.
EEG Utility in Diagnosis and Monitoring Treatment Outcomes
According to numerous studies [7,8,9,10,11,12], EEG data are also promising for the identification of depression biomarkers and for treatment prediction. Our advisors, Brainclinics, have already demonstrated that treatment based on EEG biomarkers has a significantly better outcome than treatment-as-usual (TAU) . Importantly, these promising results were achieved with simple EEG data analysis without the application of more complex modern data-science approaches. Our company plans to integrate the knowledge of EEG and psychiatry from leading neuroscientists and the potential of data analysis provided by machine learning.
The application of machine learning to study EEG signals has already shown promising results. For example, researchers from the alphabet’s experimental projects group, X (a Google sister company), spent three years developing machine-learning models to analyze EEG data . In this project (Amber), researchers demonstrated significant results in the identification of anhedonia, or the inability to feel pleasure, which is one of the main domains of depression research. However, their main goal was not just to learn how to identify and measure depression but rather to do so quickly and easily using a simple tool with only a few channels. Our company, in contrast, wants to apply their mathematical experience but to standardized EEG data with signals from 33 channels.
 Honke, G., Higgins, I., Thigpen, N., Miskovic, V., Link, K., Duan, S., Gupta, P., Klawohn, J., & Hajcak, G. (2020). Representation learning for improved interpretability and classification accuracy of clinical factors from EEG. arXiv preprint arXiv:2010.15274.