Portrait (2022-2026)

The project PORTRAIT (Improving psychiatric screening with artificial intelligence) is to develop an adaptive testing method that adjusts the questions of a test based on the subject’s previous responses. To achieve this, the project aims to extend recent advancements in recommendation systems and reinforcement learning methods to adapt tests dynamically. The project’s case study focuses on psychiatric tests, which represent a significant public health challenge. These tests currently incur high costs, but the approach developed within the project aims to reduce these costs while maintaining their reliability. A key challenge for adaptive testing is optimizing its administration to evaluate multiple psychiatric dimensions in a short period of time.

The project pursues three main objectives:

• Develop a reinforcement learning algorithm trained on data collected during psychiatric testing. This algorithm will determine, after each patient response, the next question or cognitive test to administer based on previous responses, aiming to reduce testing time.
• Evaluate this method using real psychiatric data, with validation by psychiatrists. Simpler approaches will also be implemented for comparison.
• Provide interpretability of the results to explain and justify a patient’s evaluation on psychiatric dimensions. The system must justify unanswered questions based on answered ones and the statistics obtained during the model's training phase.