Self-Evolving Human-Centered Framework for Explainable Depression Symptom Annotation

Annotation quality is critical for trustworthy mental health AI: The paper identifies that depression datasets often lack structured evidence and clear justification for diagnostic labels, which undermines both transparency and the ability to understand how AI models make decisions—a foundational problem for responsible AI in healthcare.

Expert oversight remains essential even with AI assistance: The framework combines LLM-assisted labeling with expert verification rather than automating diagnosis, reflecting the principle that AI should augment rather than replace human judgment in sensitive mental health applications.

Alignment with clinical standards improves accountability: By anchoring annotations to DSM-5-TR diagnostic criteria, the framework creates traceable connections between AI decisions and established medical standards, making it easier to audit whether the system is following recognized clinical guidelines.

Transparency through documentation enables oversight: The framework exports clinical evidence, reasoning traces, and edit histories alongside labels, providing the detailed documentation necessary for clinicians and regulators to understand and verify how the AI system arrived at its conclusions.

Iterative learning from expert feedback without retraining is promising but unproven: The dual-memory architecture is designed to incorporate expert corrections over time, but the paper acknowledges this mechanism's effectiveness "across multiple feedback cycles" remains to be evaluated—highlighting the importance of rigorous testing before deployment.