Deep Learning-Based Pain Recognition from Physiological Signals Using Hybrid CNN and Recurrent Networks

Abstract

Automatic pain recognition plays a significant role in modern healthcare systems by enabling objective and continuous monitoring of patient conditions. Traditional approaches rely on manual feature extraction from physiological signals, which requires domain expertise and may not capture complex patterns effectively. In this work, we propose a multi-level deep learning framework that performs automatic feature extraction and classification directly from physiological signals such as EEG and ECG. The proposed architecture combines Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Bidirectional Gated Recurrent Units (BiGRU) to capture both spatial and temporal dependencies in the data. The system is evaluated on a physiological signal dataset containing multiple subjects with pain intensity levels ranging from 0 to 4. Experimental results show that the baseline Random Forest model achieves 53% accuracy, while the CNN + BiLSTM model improves performance to 85%. The proposed multi-level CNN + BiLSTM + BiGRU model achieves the highest accuracy of 98%, demonstrating superior capability in distinguishing different pain levels. These results highlight the effectiveness of deep learning models over traditional machine learning approaches for pain recognition tasks.