Events

Opportunities in AI for Mental Health

Researchers are warning of a mental health pandemic: 41.5 percent of US adults and 32 percent of the adults worldwide exhibited symptoms of anxiety or depression in 2021, according to Gallup. This increase has been accompanied by spikes in suicide and “deaths of despair.” Despite the steady rise in reported mental health problems since 2011, we still lack the knowledge, tools, and human resources to provide people with the needed care—but recent advances in artificial intelligence (AI) have opened the doors for the development of therapy interventions and better predictive tools.

In this context, the topic for e-HAIL’s in-person conversation in spring 2024 was opportunities and challenges of using AI tools to address problems related to mental health. The goal of this session was twofold: (1) to introduce e-HAIL members to new mental health data sets coming on-line and (2) to identify the main research challenges in this space and brainstorm potential paths to solutions. Ultimately, thematic conversations such as this are one way in which e-HAIL fulfills its mission to facilitate new collaborations between AI and health experts to enable innovative research that advances both domains. 

e-HAIL convener Rada Mihalcea welcomed the thirty attendees from the Medical School, College of Engineering, and other U-M units, most of whom had done prior work at the intersection of AI and mental health. Amy Bohnert started the session by presenting the range of mental health and related data available to U-M faculty researchers through the Precision Health PROMPT study as well as the new data resources expected to become available with the new, larger-scale PROMPT follow-up study, COMPASS. Both the PROMPT and COMPASS studies aim to increase mental health treatment capacity through mobile health, reducing the burden on clinicians while providing individualized, around-the-clock care to patients. Bohnert and co-PI Srijan Sen have shown the positive impact of mobile technology interventions through apps that give enhanced feedback on healthy behaviors such as walking, as well as app-based therapy. While this work illustrates the promise of personalized health technologies, it also showed some of the fundamental challenges, such as appropriateness of digital interventions for different kinds of mental illnesses.

Characteristic of the ethos of e-HAIL, smaller table conversations (with lots of sticky notes) took up most of the session. Attendees broke into smaller groups to discuss challenges in mental health research and possible solutions through AI. Based on the specific interests represented at each table, the six resulting conversations each took on its own distinct flavor, capably facilitated by the e-HAIL members serving as discussion leaders (Rada Mihalcea (CoE);  

Veronica Perez-Rosas (CoE); Emily Mower Provost (CoE), David Belmonte (MM), Peter Grau (MM/VA), and COMPASS co-Is Amy Bohnert and Srijan Sen (MM). Specific research interests represented include AI assistance for counseling professionals, mood detection from speech, use of mHealth devices for mental health monitoring, precision psychiatry, clinical decision support, clinical reasoning for medical education, and more. 

Data challenges dominated many of the discussions and prompted lively exchanges of ideas and information. Researchers raised key questions around: capturing longitudinal data; creating standardized approaches to data collection and data sharing across studies; and gathering multimodal data and using AI to analyze these. An important consideration across various conversations was the difficulty and importance of ensuring that data was representative of people from diverse backgrounds. Another frequently-mentioned data-related challenge was contextualizing patient data: tying the basic data available on individuals to their personal life events such as giving birth and job loss, large scale events such as political elections, and cultural contexts relevant for better interpretation of the data.

The attendees did not lose sight of the bigger picture. They exchanged ideas about standardizing definitions about mental illnesses and tying together phenotypes with genomic data to better assess patient risk. They also highlighted the importance of seeing mental illness as a multi-stakeholder problem that involves clinicians of different specialities, patients and their families, engineers, and data scientists; any solution needs to take into account the needs and goals of each stakeholder group. 

Values and ethics around AI tools were a core concern in all table conversations. Researchers focused on the importance of having regulations around data sharing and making data sets representative of underserved populations to eliminate disparities. The risks of dehumanizing the mental health field through the creation of AI tools as well as the fraught issue of patient and clinician trust of AI-based technologies were also raised as concerns. Attention was brought to the importance of ensuring coordination between the clinical and engineering sides integrating AI tools into clinical workflows.

In the face of these challenges, the attendees brought up ongoing research projects and proposed new ideas. One suggestion was to have mobile-ready and AI-ready platforms to standardize data collection across studies and reduce the burden of creating new tools. The generation of synthetic data was proposed as a way to accelerate data collection. Mihalcea’s recently awarded grant to use AI to give feedback to counselors was pointed out as a model to improve the quality of care. Large language models were also put forward as a means to summarize large quantities of data and extract trends from unstructured data. Across the board, researchers felt that the AI’s potential to create individual, targeted treatments is a vital part of any successful approach. A final point was that an essential aspect of any solution involved guaranteeing patient privacy and control of one’s own data.

After each group presented their ideas, attendees continued discussions on their own and exchanged contact information with new potential collaborators. The lively discussions show that the diverse perspectives brought together by e-HAIL plant the seeds for developing the next generation of technology to improve (mental) health. 

Questions?

Contact J. Henrike Florusbosch, Ph.D., e-HAIL Program Manager, at jflorusb@umich.edu.