Mental health research relies on standardized diagnostic interviews (MINI, SCID) conducted by multiple interviewers, introducing significant quality biases. The lack of automated quality control tools represents a major methodological bottleneck undermining clinical research data reliability.
To develop an innovative natural language processing (NLP) tool enabling standardized and automated qualitative auditing of clinical interviews in mental health.
Our interdisciplinary approach combines psychiatric expertise, computational linguistics, and machine learning through expert knowledge acquisition, algorithm development for expert behavior emulation, corpus annotation, model training and validation, and replication across multiple questionnaires (MINI, C-SSRS, MADRS).
The project is anchored in the BIPOVITE study, a multicenter prospective observational study validating a composite medical device for bipolar disorder diagnosis. The study involves 623 adults and employs recorded questionnaires for quality assessment.
The solution aims to reduce diagnostic biases and increase the early detection of protocol deviations. The approach transforms quality control from retrospective auditing to proactive assurance, enabling 24-hour assessment. Open-source development addresses the current lack of certified tools for automated quality assessment in psychiatric clinical trials.