Mental Health World Conference 2026

Abstract

Postnatal depression (PND) is a major global mental health challenge that often remains unrecognized, affecting the well-being of mothers, infants, and families. Early identification is essential, yet current screening approaches rely largely on selfreported questionnaires and may overlook subtle behavioural and emotional changes. This study presents a prototype Oriented Multimodal Explainable Artificial Intelligence (OMEAI) framework designed to support early PND clinical screening by integrating facial behavioural cues, vocal characteristics, textual sentiment, and Edinburgh Postnatal Depression Scale (EPDS) scores within a single explainable framework. The system combines multimodal data fusion with explainable artificial intelligence (AI), reliability & validation layer, and SOAP-based clinical reporting to provide transparent and clinically meaningful screening outputs.

The framework was evaluated using speech and language data from the Extended DAIC WOZ dataset, a prototype facial image dataset comprising 100 representative postnatal facial samples and one multimodal postpartum case integrating facial, vocal, textual, and psychometric information. Compared with previous single-modality studies using the Extended DAIC WOZ dataset, the multimodal framework demonstrated improved clinical screening performance, with the receiver operating characteristic area under the curve (ROC AUC) increasing from 0.84 to 0.92, while also providing interpretable explanations to support clinical understanding. The inclusion of explainability and reliability measures strengthened confidence in the models outputs and highlighted the complementary value of combining multiple behavioural and clinical indicators.

These preliminary findings suggest that the OMEAI framework could support earlier identification of women at risk of postnatal depression by combining multiple sources of clinical information in a transparent way. Although further validation in larger postpartum populations is needed, the framework shows promise as a clinical screening support tool that may complement routine maternal mental health screening rather than replace clinical judgement.