A CrossMod-Transformer deep learning framework for multi-modal pain detection through EDA and ECG fusion

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Pain Medicine

Background:

• Objective pain assessment is crucial for effective pain management and healthcare efficiency.
• Accurate pain evaluation is challenging due to subtle physiological and behavioral indicators and individual variability.
• Automatic pain assessment systems offer technological solutions to enhance pain evaluation processes.

Purpose of the Study:

• To propose a novel multi-modal ensemble deep learning framework for automatic pain recognition.
• To combine electrodermal activity (EDA) and electrocardiogram (ECG) signals for objective pain assessment.
• To evaluate the generalisability and performance of the proposed framework on independent datasets.

Main Methods:

• Developed a uni-modal approach (FCN-ALSTM-Transformer) integrating Fully Convolutional Network, Attention-based LSTM, and Transformer blocks.
• Introduced a multi-modal approach (CrossMod-Transformer) using a dedicated Transformer architecture to fuse EDA and ECG signals.
• Conducted experiments on the BioVid dataset and cross-dataset validation using the AI4PAIN 2025 dataset.

Main Results:

• The CrossMod-Transformer achieved 87.52% accuracy on the BioVid dataset.
• The model demonstrated strong generalisability with 75.83% accuracy on the AI4PAIN dataset.
• The proposed framework outperformed several state-of-the-art uni-modal and multi-modal methods.

Conclusions:

• The novel multi-modal deep learning framework shows significant potential for reliable automatic pain recognition.
• The CrossMod-Transformer enhances objective pain assessment by effectively fusing physiological biosignals.
• This research supports the development of more objective and inclusive clinical pain assessment tools.