Explainable AI for Screening Abuse-Related Trauma in Bangladeshi Children: A Training-Free Multimodal Framework Evaluated on Noise-Aware Synthetic Data

Addresses a critical gap in mental health infrastructure: Bangladesh has only six child psychiatrists nationwide and 1.17 mental-health professionals per 100,000 population, with no existing Bengali-language screening tool for abuse-related trauma in children—ShishuRaksha AI was designed to help fill this urgent need.

Uses explainability as a core feature, not an afterthought: Every risk score includes clinically weighted explanations showing which factors (questionnaire responses, drawings, facial expressions, or narrative text) contributed to the assessment, helping clinicians understand the AI's reasoning rather than treating it as a "black box."

Combines multiple assessment methods to improve accuracy: The framework integrates four different screening modalities (validated questionnaires, Bengali narratives, drawings, and facial affect), achieving an AUC of 0.874 compared to 0.756 for questionnaires alone, suggesting that multimodal fusion can meaningfully improve detection.

Designed with built-in safeguards and transparency about limitations: The tool is explicitly positioned as a decision-support aid (not a diagnostic tool), includes referral routing to national child-protection services, and the researchers openly acknowledge that their evaluation used only synthetic data with no real clinical dataset—being transparent about what the system can and cannot do.

Culturally adapted for local context: Reports are generated in both Bangla and English, and the framework incorporates culturally grounded priors (such as for interpreting House-Tree-Person drawings), recognizing that screening tools must be tailored to the populations they serve.