Abstract
According to the WHO, the number of mental disorder patients, especially depression patients, has overgrown and become a leading contributor to the global burden of disease. With the rising of tools such as artificial intelligence, using physiological data to explore new possible physiological indicators of mental disorder and creating new applications for mental disorder diagnosis has become a new research hot topic. We present a multi-modal open dataset for mental-disorder analysis. The dataset includes EEG and recordings of spoken language data from clinically depressed patients and matching normal controls, who were carefully diagnosed and selected by professional psychiatrists in hospitals. The EEG dataset includes data collected using a traditional 128-electrodes mounted elastic cap and a wearable 3-electrode EEG collector for pervasive computing applications. The 128-electrodes EEG signals of 53 participants were recorded as both in resting state and while doing the Dot probe tasks; the 3-electrode EEG signals of 55 participants were recorded in resting-state; the audio data of 52 participants were recorded during interviewing, reading, and picture description.
Introduction
Diagnosing Major Depressive Disorder (MDD) is currently a subjective and labor-intensive process. It relies heavily on clinical interviews and self-reporting scales, which lack the objective physiological "biomarkers" found in other illnesses like hypertension or diabetes. While researchers aim to use artificial intelligence to identify these biomarkers, high-quality data is notoriously difficult to acquire. Depressed patients often struggle with the motivation required for long lab experiments, and data must be collected before medication starts to avoid confounding brain activity signals. The Multi-modal Open Dataset for Mental-disorder Analysis (MODMA) addresses this scarcity by providing a comprehensive, multi-modal dataset to the public.
Methods and Experimental Paradigm
The researchers employed a three-pronged data collection strategy to capture different physiological and behavioral facets of depression:
- High-Density EEG: Using a traditional 128-electrode elastic cap to capture "gold standard" brain activity.
- Pervasive EEG: Using a novel, wearable 3-electrode collector targeting the prefrontal lobe (Fp1, Fpz, and Fp2). This device is designed for real-world, "in-the-wild" pervasive computing applications.
- Audio Recordings: High-fidelity speech data captured via professional-grade microphones during structured tasks.
The experimental paradigm included both resting-state recordings and a Dot-probe task. In the Dot-probe task, participants view pairs of emotional (sad, fear, happy) and neutral faces. Their reaction times and event-related potentials (ERPs) help quantify "attention bias"—the tendency of depressed individuals to struggle with disengaging from negative stimuli.
The MODMA Dataset
The MODMA dataset includes data from participants carefully diagnosed by professional psychiatrists:
- 128-electrode EEG: 53 participants (24 MDD, 29 Healthy Controls) recorded during resting-state and Dot-probe tasks.
- 3-electrode EEG: 55 participants (26 MDD, 29 Healthy Controls) recorded during resting-state.
- Audio Data: 52 participants (23 MDD, 29 Healthy Controls) performing three tasks: an 18-question interview, reading a short story ("The North Wind and the Sun"), and describing emotive pictures.
The data is provided in the standardized BIDS (Brain Imaging Data Structure) format, facilitating integration with existing analysis pipelines.
Key Insights
- Pervasive Potential: Validation shows that even the minimal 3-electrode setup can distinguish between MDD and healthy controls with significant accuracy, suggesting high potential for daily-life monitoring.
- Multi-modal Fusion: While individual modalities are useful, "feature-level fusion" (combining EEG and audio) yields the highest diagnostic accuracy.
- Hemispheric Asymmetry: EEG signals from the right hemisphere were found to be more distinctive in depression, supporting the theory of frontal EEG asymmetry as a potential biomarker.
- Speech as a Biomarker: Changes in vowel formants and articulatory coordination in MDD patients provide a non-invasive window into their mental state.
Validation Results
Several studies performed on this dataset demonstrate its utility:
- EEG Classification: Using 128-channel resting-state data, researchers achieved 82.31% accuracy in identifying MDD.
- Wearable Success: The 3-channel wearable EEG achieved 72.25% accuracy, which increased to 86.98% when fused with audio-stimulated EEG features.
- Speech Recognition: Audio-based models reached 75.8% accuracy for males and 68.5% for females, highlighting gender-dependent differences in depression expression.
Limitations and Future Work
The primary limitation is the sample size (approximately 53-55 participants), which is small for deep learning but standard for clinical EEG studies. Future work involves expanding the dataset to include more participants and other disorders such as schizophrenia and anxiety. The authors also plan to incorporate eye-movement tracking and MRI data to create a more robust multi-modal profile.
Summary updated