I am honored to be the co-first author of this paper, and my main contrbution are data extraction and visualization.
Based on the 2019-2024 scoping review of 95 peer-reviewed articles, this study mapped the landscape of large language models (LLMs) applications in mental health across three key domains. The analysis revealed that LLMs are predominantly utilized for screening and detection of mental disorders (71%), with particular emphasis on depression detection (35%) and suicide risk prediction (13%). Additionally, LLMs demonstrate significant potential in supporting clinical treatments (33%) and facilitating mental health counseling and education (12%). Comparative assessments indicate that LLMs exhibit superior capabilities in information processing and natural language response generation relative to traditional non-transformer models and human performance in specific contexts. The research identified distinct advantages among different LLM architectures for various mental health applications, highlighting their promising role in addressing critical challenges in global mental healthcare, including detection efficiency, treatment effectiveness, privacy protection, and access to specialized care. These findings provide essential scientific evidence for the development and implementation of LLM-enhanced mental health interventions, which may significantly improve early detection rates and expand access to mental healthcare resources.