TriMind

A Large-Scale Three-Electrode EEG Dataset for Mental Disorder Differential Diagnosis

About TriMind

TriMind is a three-lead frontal EEG dataset for multi-psychiatric disorder recognition and differential diagnosis, collected with a lightweight portable setup using only Fp1, Fpz, and Fp2.It includes four subject groups: major depressive disorder (MDD), anxiety disorder (ANX), schizophrenia (SCH), and healthy controls (HC).Unlike most existing EEG datasets focused on single-disorder or binary classification, TriMind supports four-class, three-class, and binary classification settings for studying both shared and disorder-specific neural patterns.The dataset contains both resting-state EEG and emotional auditory stimulation EEG, providing complementary information on stable and task-evoked brain activity.Overall, TriMind comprises 729 subjects and about 170,586 seconds of EEG recordings, offering a practical benchmark for low-channel wearable psychiatric EEG research.

Experimental Paradigm

The experimental paradigm of TriMind consists of both resting-state recording and emotional auditory stimulation recording. During the resting-state stage, subjects remained relaxed and quiet for a total of 90 seconds to capture relatively stable background neural activity. During the stimulation stage, subjects were exposed to different types of emotional auditory stimuli for a total of 144 seconds, in order to evoke dynamic brain responses related to emotional processing. Specifically, the experiment was organized into fixed 6-second units, in which resting periods alternated with positive, neutral, and negative auditory stimulation segments, enabling the EEG signals to reflect both intrinsic baseline neural characteristics and response patterns under different emotion-inducing conditions.

Comparison with other datasets

Compared with most existing EEG datasets that focus on single-disorder recognition or binary classification against healthy controls, TriMind offers greater diversity in classification tasks. It supports not only four-class classification across major depressive disorder, anxiety disorder, schizophrenia, and healthy controls, but also a variety of three-class and binary classification settings. This design facilitates the study of both shared abnormalities and disorder-specific differences across psychiatric conditions, and better reflects the practical needs of differential diagnosis in real clinical scenarios.

Various visual forms of EEG

a) Time-Voltage EEG

b) Time-Frequency EEG

c) PSD EEG

Access

Follow the steps below to request access to the TriMind dataset.

→

Receive Dataset

Please contact us by email to request access to the dataset (contact: 2025170828@mail.hfut.edu.cn).

After we receive your request, we will send the dataset to you.

You can download the sample data below.

🗂 Download Sample Data

Publications

TriMind: A Large-Scale Three-Electrode EEG Dataset for Mental Disorder Differential Diagnosis

Electroencephalography (EEG) is a promising neural signal for mental disorder assessment. However, existing EEG datasets mainly focus on single-disorder settings, limiting their use in realistic mental disorder differential diagnosis. To address this gap, we present a large-scale three-electrode EEG dataset for multi-disorder analysis, covering depression, anxiety and schizophrenia, with all labels clinically confirmed by professional psychiatrists. The dataset contains 729 subjects with both resting-state and stimulation-state EEG recordings, and enables flexible differential diagnosis tasks under arbitrary disease combinations. On top of this benchmark, we propose a prior-guided self-supervised pretraining framework for mental disorder differential diagnosis. By introducing prior-guided augmentation and prior-guided masking, the proposed method exploits physiological priors to mine richer disorder-discriminative cues from limited EEG electrodes. Extensive experiments verify the effectiveness of the proposed method. The dataset and code are publicly available at https://wushidiyishenqing.github.io/dataset-webpage/.

📄 Paper (PDF) 🛠 arXiv (Coming soon) 🐙 Code