Abstract

Depression is a major mental-health condition that affects emotion, behaviour, communication and daily functioning, and its early identification remains a significant challenge for both clinicians and researchers. Conventional assessment relies mainly on clinical interviews and self-report instruments such as the PHQ and BDI, which are subjective and dependent on the willingness and ability of an individual to accurately report symptoms. Advances in Generative Artificial Intelligence (GenAI), deep learning and multimodal data fusion have opened new possibilities for analysing text, audio, video and behavioural signals together to support early depression-risk detection. This paper presents a structured review of recent literature on multimodal and Generative-AI-based approaches for depression detection, tracing the progression from traditional machine-learning techniques to deep-learning, transformer-based and generative architectures. Eight representative studies are analysed and compared in terms of methodology, key findings and limitations. Based on this synthesis, key research gaps are identified, including limited integration of GenAI with rich multimodal information, scarcity of large and diverse real-world datasets, generalization difficulties, GenAI reliability concerns and privacy and ethical issues associated with sensitive mental-health data. A conceptual multimodal GenAI framework that combines text, audio, video and behavioural indicators while addressing reliability, privacy and explainability is subsequently outlined as a direction for future work.

Keywords

How to Cite This Article

Rights