SBIR Phase II: An Intelligent Mental Health Therapy System

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Small Business Innovation Research, Phase II project is to help make therapy more consistent with patient preferences, beliefs, and values to maximize engagement in therapy and improve patient outcomes. Therapy for mental health problems is highly effective, yet many patients drop out before getting the full benefit because they are not satisfied or engaged in the therapy. The proposed project involves collecting data on all of patients actions in the online treatment system along with their ratings of each activity and their symptom improvement over time. The research and development team will use this data to create a machine learning system that will make suggestions for best next steps in therapy based on what thousands of other users experienced. This is the intelligent counseling system. It will work very similarly to movie streaming services or online book sellers who recommend movies or books to you based on your past preferences and the preferences of thousands of other users.The proposed project will develop a feedback and recommendation system based on advanced analytics and machine learning techniques to provide personalized treatments to customize and individualize online mental health treatment, the Intelligent Counseling System (ICS). This personalized system will contain a number of alternative treatment items from several theoretical perspectives, using a variety of patient interactive activities, varying in format, length, pace, and other characteristics. In such a setting, a recommendation system can predict the users' preferences and recommend the subsequent treatment component. In addition, to achieve maximum adherence and to decrease the attrition rate, the platform will enable personalized motivational interventions and supportive messaging. The delivery times and the content of supportive messaging will adapt and vary depending on the projected treatment progress. Our machine learning based system will be trained incrementally as more data becomes available over time, thus it will benefit from improved accuracy over time. We will extract local, semi-local, and global temporal features at multiple temporal resolutions and will use feature selection techniques to identify which factors contribute to the success of treatments for patients, and to predict if a user is improving or is deteriorating. This will result in adaptive motivational messages and recommendation for tailoring treatment in term of important identified treatment features.