A Comprehensive Review of Datasets for Clinical Mental Health AI Systems

Aishik Mandal, Prottay Kumar Adhikary, Hiba Arnaout, Iryna Gurevych, Tanmoy ChakrabortyarXiv 2025 · DOI · arXiv

Abstract

Mental health disorders represent a growing global challenge, yet the development of AI-driven clinical tools is hindered by a significant data bottleneck. This paper provides the first comprehensive review of 89 real-world clinical mental health datasets and 16 synthetic data benchmarks. By establishing a multi-dimensional taxonomy, the authors move beyond unvalidated social media proxies toward high-quality, clinically grounded AI systems, identifying critical gaps in accessibility, modality, and sociocultural representation.

Introduction

The gap between the rising demand for mental health services and the availability of trained clinicians is widening. While Artificial Intelligence (AI) holds promise for scalable diagnosis and intervention, it faces a "data bottleneck." Early research often relied on social media data (e.g., Reddit, Twitter), which lacks clinical validation. Conversely, actual clinical data—such as therapy transcripts, EHRs, and brain scans—is frequently restricted due to privacy regulations like GDPR and HIPAA. This paper addresses this fragmentation, providing a roadmap for researchers to navigate high-quality, clinically validated resources.

Methodology: A Multi-Dimensional Taxonomy

To organize the complex landscape of clinical data, the authors conducted a systematic review across major databases, filtering 560 initial records down to 89 core clinical datasets. They developed a taxonomy to categorize these resources based on disorder type, data modality, task type, accessibility, and sociocultural context.

Taxonomy of clinical mental health datasets

This framework allows for a granular analysis of how datasets align with gold-standard clinical tools, such as the PHQ-9 for depression and the PANSS for schizophrenia.

The analysis reveals a field in transition. While clinical datasets provide the "ground truth" necessary for medical trust, they are often smaller and less accessible than social media proxies. Research is heavily skewed toward Schizophrenia and Depression, while Bipolar Disorder and Anxiety remain significantly underrepresented in publicly available clinical data.

Distribution of clinical datasets by disorder, access, and geography

Privacy remains the primary gatekeeper; the majority of clinical data is "Private" or "Restricted," creating a reproducibility crisis. Furthermore, there is a glaring "WEIRD" (Western, Educated, Industrialized, Rich, and Democratic) bias. Most datasets originate from the US, China, or Europe, leaving a void of data from the Global South where mental health symptoms may be expressed through different cultural idioms.

The Rise of Synthetic Data

Synthetic data generation using Large Language Models (LLMs) is an emerging frontier to bypass privacy hurdles. The authors reviewed 16 synthetic datasets, such as MentalChat16k, which simulate patient-clinician interactions. While promising for training text-based models without risking patient privacy, current synthetic data lacks the multimodal depth (audio and video) required for comprehensive clinical assessments.

Future Directions and Conclusion

The review identifies several critical gaps that must be addressed to advance the field:

  1. Longitudinal Data: Most datasets are snapshots; longitudinal data is needed to predict relapses.
  2. Standardization: The absence of an "ImageNet for Mental Health" makes model comparison difficult.
  3. Multimodal Privacy: Anonymizing video and audio without losing clinical features remains a technical challenge.

The authors advocate for Federated Learning and more sophisticated multimodal synthetic data to bridge these gaps and create more equitable, robust AI systems for mental healthcare.

Summary updated

Abstract

Mental health disorders are rising worldwide. However, the availability of trained clinicians has not scaled proportionally, leaving many people without adequate or timely support. To bridge this gap, recent studies have shown the promise of Artificial Intelligence (AI) to assist mental health diagnosis, monitoring, and intervention. However, the development of efficient, reliable, and ethical AI to assist clinicians is heavily dependent on high-quality clinical training datasets. Despite growing interest in data curation for training clinical AI assistants, existing datasets largely remain scattered, under-documented, and often inaccessible, hindering the reproducibility, comparability, and generalizability of AI models developed for clinical mental health care. In this paper, we present the first comprehensive survey of clinical mental health datasets relevant to the training and development of AI-powered clinical assistants. We categorize these datasets by mental disorders (e.g., depression, schizophrenia), data modalities (e.g., text, speech, physiological signals), task types (e.g., diagnosis prediction, symptom severity estimation, intervention generation), accessibility (public, restricted or private), and sociocultural context (e.g., language and cultural background). Along with these, we also investigate synthetic clinical mental health datasets. Our survey identifies critical gaps such as a lack of longitudinal data, limited cultural and linguistic representation, inconsistent collection and annotation standards, and a lack of modalities in synthetic data. We conclude by outlining key challenges in curating and standardizing future datasets and provide actionable recommendations to facilitate the development of more robust, generalizable, and equitable mental health AI systems.
BibTeX
@misc{mandal2025comprehensive,
  author = {Aishik Mandal and Prottay Kumar Adhikary and Hiba Arnaout and Iryna Gurevych and Tanmoy Chakraborty},
  title = {{A Comprehensive Review of Datasets for Clinical Mental Health AI Systems}},
  year = {2025},
  eprint = {2508.09809},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.09809},
}
Figure