This paper explores the potential of artificial intelligence (AI) to develop objective, scalable, and timely diagnostic tools for major depressive disorder (MDD), a leading cause of disability worldwide. Given that the diagnosis of this disorder still relies heavily on subjective clinical assessments, this paper explores the potential of AI-powered approaches to detect and diagnose depression. Based on a systematic review of 55 key studies, we conduct a comprehensive survey of state-of-the-art AI methods for depression detection and diagnosis. We introduce a novel hierarchical taxonomy that structures the field by diagnostic versus predictive approaches, data modality (text, speech, neuroimaging, multimodal), and computational model class (e.g., graph neural networks, large-scale language models, hybrid approaches). Our in-depth analysis reveals three key trends: the rise of graph neural networks for brain connectivity modeling, the rise of large-scale language models for language and conversational data, and a growing focus on multimodal convergence, explainability, and algorithmic fairness. Along with methodological insights, we provide an overview of key public datasets and standard evaluation metrics as a practical guide for researchers. By synthesizing current advances and highlighting open challenges, this paper offers a comprehensive roadmap for future innovations in computational psychiatry.
