Related Experiment Video
Updated: May 21, 2025
05:51
Exploring the Use of Isolated Expressions and Film Clips to Evaluate Emotion Recognition by People with Traumatic Brain Injury
Published on: May 15, 2016
8.9K
MT-RCAF: A Multi-Task Residual Cross Attention Framework for EEG-based emotion recognition and mood disorder
Xinni Kong1, Yaru Guo1, Yu Ouyang1
1School of Computer Science and Technology, Hangzhou Dianzi University, Hangzhou, Zhejiang, China.
Computer Methods and Programs in Biomedicine
|May 19, 2025
Summary
This study introduces a novel Multi-Task Residual Cross Attention Framework (MT-RCAF) to improve emotion recognition and mood disorder detection by analyzing their relationship using EEG data.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Psychiatry
Background:
- Prolonged abnormal emotions can lead to mood disorders like anxiety and depression.
- Current EEG-based emotion and mood disorder detection methods are often studied in isolation, missing potential synergistic benefits.
- Understanding the link between emotions and mood disorders is crucial for identifying causes and developing interventions.
Purpose of the Study:
- To investigate the relationship between emotional states and mood disorders using electroencephalogram (EEG) data.
- To propose and validate a novel Multi-Task Residual Cross Attention Framework (MT-RCAF) designed to enhance both emotion recognition and mood disorder detection.
- To leverage shared and task-specific features for improved classification performance in both domains.
Main Methods:
- Developed the Multi-Task Residual Cross Attention Framework (MT-RCAF) incorporating Feature Extraction, Residual Multi-head Cross Attention (RMCA), Gated Multi-embedding (GME), and Task Tower Classification modules.
- The RMCA module dynamically captures shared and task-specific information through attention weights.
- The GME module filters irrelevant features, while the Task Tower Classification module balances task losses.
Main Results:
- Experiments on the DEAP and EMDD datasets demonstrated MT-RCAF's effectiveness.
- Significant accuracy improvements were observed: 3.22% for emotion recognition and 3.91% for mood disorder detection in strongly correlated tasks.
- Generally correlated tasks showed average accuracy increases of 2.87% for valence and 3.34% for arousal.
Conclusions:
- The study confirms a deep-learning-validated relationship between emotions and mood disorders.
- Interconnected task learning within the MT-RCAF framework leads to more accurate and robust classification results.
- Findings suggest mood disorders heighten sensitivity to negative emotions, and intense emotions aid mood disorder detection.
