Mental Health Data Needed for AI

Artificial Intelligence (AI) in Mental Health Sciences broadly represents the application of Machine Learning methods in the analysis of mental health data. These efforts have experienced recent expansion, with more than 300 studies occurring between 2015 and 2019 — more than all previous years combine. Most studies in this segment, aside from cognitive impairment, focus on depression. Schizophrenia, suicide ideation/attempts, general mental health, medication adherence, and uni/bi-polar disorders are also considered. The data sources leveraged include surveys, biomarkers, medical imaging (MRI, fMRI, EEG), motion video, and Electronic Health Records (EHR), as well as smartphone and social-media records.

The models constructed are mostly Supervised Machine Learning (SML) performing binary classification, with percent-accuracy reaching high 90s for clinical measurements and in the 60s for smartphone/social media records. Unsupervised Machine Learning (UML) has been utilized to classify physiological subtypes of depression and schizophrenia with medical imaging. Lastly, Natural Language Processing (NLP) has been leveraged to identify, in automated fashion, the patients experiencing severe mental illness from their EHR. In many cases, Neural Networks (NN) are capable of accurately interpreting far more complex data than the more traditional ML algorithms.

However, for Mental Health Sciences to leverage them, much work is yet to be done creating large, centralized data repositories harboring data on both the diverse types and sources of mental health information. Such an effort could likely originate with guidance and standards from the National Institute of Mental Health (NIMH/NIH) and be loosely modeled after the Genomic Data Commons (NCI) and/or the US's Human Samples and Data Repository - Clinical Trial. Such guidance could anticipate the directions of development in NLP that could optimize data preparation for NN modeling. The long-term benefits to such an effort could be the development of a repository for large numbers of bio-phyco-social profiles that could be mined for helpful mental health information, such as early disease identification, disease prevention, factors protective against development of disease, and general mental health optimization.

NN(AI) models are ready for application in the Mental Health Sciences space. To realize their potential, appropriate data set preparation may need to be the near-term focus. 

Content summarized from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7274446/.