eXplainable AI in Personalized Mental Healthcare

This research collaboration aims to develop novel decision support tools for personalized online therapy by combining eXplainable Artificial Intelligence (XAI) and expert knowledge from mental healthcare professionals. Its main goal is to assist therapists during their in-between session slots by indicating which clients require close examination in their data that might indicate a need for intervention.

Background

NiceDay is fully committed to deliver an evidence-based service and is fully aware of the multiple advantages that user data can provide in making treatments more personalized and effective. One approach to this is to help therapists in their treatment workflow and decision making process. Together with them we are deciphering which parts of treatment can be assisted with the introduction of new technologies.

This MIT R&D collaboration project proposes to advance the explainability paradigm in AI in the mental healthcare sector through the development of novel eHealth technologies aiming to support the work of therapists in a data-driven approach. NiceDay, Deeploy, and Councyl have partnered to make this challenge possible. 

By leveraging NiceDays’ current online mental healthcare service available for therapists and clients, AI models will contribute to determine the optimal treatment practices. In particular, we aim to assist therapists during their in-between session slots by indicating which clients require close examination in their data that might indicate a need for intervention.

The AI expert partners, Councyl and Deeploy, support NiceDay in the development of accurate, human-in-the-loop AI models, as well as methods to make these models explainable, accountable and trustworthy. This will lead to the responsible application of AI, leading to more effective therapists and better service for clients. We expect to become one of the first mental healthcare platforms that helps professionals make choices during the course of treatment through AI algorithms with feedback loops. 

In the following diagram, a high level overview of how the services of the three partnering organizations interact in order to deliver the proposed XAI-based decision support tool.  

Target client group

A specific client group will form the basis for the initial proof of concept, selected based on defined scoping criteria such as diagnosis, severity, personality type, engagement levels with the platform, availability of validation data, decision-making trade-offs, and educational background. This targeted approach ensures the development of a tailored AI solution that addresses the nuanced needs of a particular client segment, setting the stage for broader application across diverse client profiles.

Expected results

The proposed results of  this project are:

1. A prototype of 3 systems, that are connected through APIs and feedback loops, for:
   1. The design and development of decision support algorithms for recommender systems through continuous monitoring of the choice behavior of experts and sensor data.
   2. The deployment, management and explainability of productionized recommender systems (MLOps for Incremental Learning).
   3. The daily operation of recommender systems , which predicts the best action for each client given its personal information and historical treatment.
2. Novel explainability methods in order to explain the predictions to therapists. The explainability methods come along with models and are deployed on the same system. There are multiple unique XAI methods which need to be researched and explained.
3. An incremental learning method which translates feedback on outcomes and explanations to improvements of the model. 
4. A novel approach to transparently measure the effect of supportive ML algorithms on expert decision making, using Behavioural AI Technology; called quantifying the human-in-the-loop.
5. Validated ML models that are used within NiceDay to improve the mental healthcare industry and assist both experts and patients, by recommending the best action for the therapist to take, given a patient’s state and history.

Partners

NiceDay is a leading mental healthcare platform, providing a service to therapists to help treat clients going through mental difficulties more effectively, ultimately strengthening the mental wellbeing of as many people as possible and aiming to make healthcare sustainable.

Deeploy has developed a platform for data scientists to apply AI responsibly, making AI decisions explainable and traceable for everyone. Deeploy has a particular focus on healthcare, as the potential value and social impact of AI is often enormous, but can only be reached if we make sure that everyone understands how decisions are made.

Councyl helps organizations manage repetitive professional decisions, through a software platform that structures, models, and controls these decisions. With its unique Behavioural AI Technology (BAIT), Councyl makes implicit expert trade-offs explicit in fully transparent choice models, without requiring historical data. The choice models form the basis to monitor, steer and improve the decision-making process without compromising quality and efficiency.

Research duration

Start date: 1-Apr-2024
End date: 1-Apr-2026