This page was translated automatically from Turkish. Read the original

This page is an AI-assisted Turkish summary of the study presented at the 26th World Congress of Psychiatry. The study belongs to the presenting authors; please refer to the original source when citing it.

A regulatory framework for continuously learning artificial intelligence in monitoring mood disorders: MoodMon certification under the European Medical Device Regulation

REGULATORY FRAMEWORK FOR CONTINUOUSLY LEARNING AI IN AFFECTIVE DISORDER MONITORING: MOODMON CERTIFICATION UNDER THE EUROPEAN MEDICAL DEVICE REGULATION (ID 3910)

The AEGIS framework, based on predefined change control criteria, has demonstrated that continuously learning artificial intelligence can be certified in compliance with the MDR and was accepted by the Notified Body as part of MoodMon's Class IIa submission.

The aim is to develop a regulatory approach that ensures continuously learning artificial intelligence remains compliant with the European Medical Device Regulation (MDR) when used to monitor mood disorders. The poster identifies a regulatory dilemma: under the MDR, each retraining cycle may be considered a major modification requiring full recertification, making adaptive artificial intelligence effectively unfeasible in certified devices. The AEGIS (Adaptive Engine Governance and Integrity Safeguards) framework establishes predefined change control criteria for each retraining cycle and aligns with the Predetermined Change Control Plan (PCCP) concept under the EU MDR and the US FDA's approach to AI-enabled devices. The cycle consists of four steps: collecting clinician feedback and new patient data; periodically retraining population-level and patient-specific models; checking algorithmic stability against sensitivity and specificity thresholds derived from clinical study data (updates that remain within these limits are considered minor changes and proceed, while those outside the limits fall beyond the scope of the criteria); and conducting post-market surveillance to verify that retraining results remain within the predetermined performance envelope. The clinical basis is a study involving 75 patients with bipolar disorder (BD): across the full dataset, sensitivity was 86.6% and specificity was 98.59%; among patients who submitted at least 50% of the expected voice recordings, sensitivity increased to 95.38%, while specificity remained at 98.6%. The AEGIS framework was submitted as part of MoodMon's Class IIa certification dossier and accepted by the Notified Body. The authors state that this framework reconciles the need for adaptive personalization with MDR certification requirements, that regulatory barriers disproportionately affect underserved regions where psychiatric workforce shortages are greatest, and that a clear certification pathway supports equitable access across geographic and socioeconomic groups.

All WCP 2026 summaries