Automated Mental Health Assessment (AMHA) to Predict Destructive Behaviors
Abstract
This proposal will research and engineer the components needed for an automated mental health assessment system (AMHA), using advances in computational behavior and artificial intelligence (AI) that are based on validated neuroscience. This research will test how well variables that quantify approach/avoidance decisions by individuals can be used to predict pastand current destructive behaviors (DB) and mental health (MH) problems. It will use a recently discovered engineering-based framework for approach/avoidance decision making, referred to as relative preference theory (RPT) to predict MH/DB. DB and MH include a range of issues that candisrupt mission readiness, mission execution, and post-mission assessment, which are nowrecognized as significant challenges for the DoD and psychiatry and can be co-morbid with DB and exacerbate the risk of its incidence. Currently no system exists that can rapidly and automatically assess the probability of DB and separate them from the broad array of potentially co-morbid MH problems in naval personnel, other members ofthe DoD, or civilians. Tools that can assess DB and MH are needed for one-time assessments, for longitudinal assessment of the course of risk, and for assessing the rapidity with which intervention is needed. Such tools are needed in the context of an increasing incidence of DB and MH problemswithin the Navy and other military services over the past decades, and reports of increasing first responder MH and DB concerns in the context of national emergencies such as the Covid-19 pandemic. The technical approach proposed for this problem follows a novel path suggested bythe AI literature, that has not been implemented by existing frameworks using: (i) text mining, (ii) surveys at big data scale (including ecological momentary assessments), or (iii) smartphone/wearable sensors. This new approach focuses on using computational behavior (i.e., advanced modeling with mathematical cognitive science) as the feature set that is used as input forclassification by data-driven AI. This new approach uses cutting-edge research out of MGH/Harvard and Northwestern that has identified a mathematical function space for approach/avoidance decisions that meets Feynman criteria for lawfulness, and, as a feature set, shows very strong prediction of DB. This function space for approach/avoidance decisions is based on RPT, and has been validated against human brain imaging and psychiatric neuroscience. RPTintegrates reward/aversion frameworks behind two Nobel Prizes in Economics, and the core frameworks used in neuroscienchesia). This work will study how well RPT variables predict the full set of DB and psychiatrically recognized MH problems. It will further test which natural language processing (NLP) variables accurately produce RPT preferencecurves and unique RPT trait variables in individuals. It will then quantify the false positive and false negative rates for NLP variables, in an RPT framework, predicting DB/MH. This work will further test the integration and scalability of these functions as the first AMHA, one that could run in real time without needing access to medically protected information in individuals. By completing this work, this proposal will both produce the necessary research for an AMHA, andproduce a working prototype for follow-up implementation science. All the necessary p probability of success. Furthermore, the lead investigators forthis team are two internationally recognized researchers, one a pioneer in AI, and the other a pioneer in reward/aversion neuroscience who have a long history of success with government funded research.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 15, 2021
- Source ID
- N000142112216
Entities
People
- Hans Breiter
Organizations
- Northwestern University
- Office of Naval Research
- United States Navy