Explainable Automated Pain Level Classification Using a Deep Neural Network with EEG Signals

Abstract Background Traditional and handcrafted approaches often neglect the spatial distribution and temporal dynamics of raw EEG signals, creating a gap in objective pain classification and clinical trust. As a result, existing pain classification systems overlook important neural patterns associated with pain perception, limiting both classification performance and clinical interpretability. Objectives In this paper, we introduce a novel electroencephalography (EEG)-based approach to develop a specialised Spatial-Temporal Convolutional Neural Network (PainNet-ST) that integrates an Explainable AI (XAI) method to establish a three-class metric model for pain severity classification (low, medium, and high). Methods Our method includes the conversion of the raw EEG data from the OpenNeuro ds005280 dataset (nociceptive-evoked potentials) into multi-spectral topography maps (2D images) to preserve spatial distribution and temporal dynamics. Our proposed PainNet-ST model integrates spatial-temporal decoupling, Bayesian optimisation (implemented via the Optuna approach), and dual-layer XAI. Our analysis focused on the neural feature representations across the somatosensory cortex and the N2-P2 temporal window. Proposed PainNet-ST model was benchmarked against Support Vector Machines (SVM), Random Forest (RF), 2D-CNN, and STAM-net. For explainability, SHAP-based visualisation was incorporated to address the gap in clinical trust. Results Our proposed PainNet-ST model was evaluated against the benchmarked models with 11 performance metrics using accuracy, precision, recall, specificity, F1-score, False Positive Rate (FPR), False Negative Rate (FNR), Negative Prediction Value (NPV), False Discovery Rate (FDR), Cohen’s Kappa (k), and Area Under the Curve (AUC), with results showing that the proposed model PainNet-ST was better than the other benchmark models in every classification task. That is, in this three-class pain classification problem, an accuracy of 92.28, F1-score of 92.29%, precision of 92.35%, recall of 92.28%, and specificity of 96.14% were obtained. Conclusion In the multiclass raw EEG-based pain classification task (low, medium, high), the model reached 92.28% accuracy, demonstrating the efficacy of the proposed method for the classification of pain. The proposed PainNet-ST model successfully addresses multi-class raw EEG-based pain classification by integrating spatial-temporal decoupling, Bayesian optimisation, and dual-layer XAI.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-16
DOI
https://doi.org/10.21203/rs.3.rs-11448439/v1
Primary Topic
EEG and Brain-Computer Interfaces
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Explainable Automated Pain Level Classification Using a Deep Neural Network with EEG Signals

Salvin Prasad, Jeffrey Soar, Aruna Devi, Ravinesh Deo et al.
3 links
Research Square
EEG and Brain-Computer Interfaces
preprint

Explainable Automated Pain Level Classification Using a Deep Neural Network with EEG Signals

Salvin Prasad, Jeffrey Soar, Aruna Devi, Ravinesh Deo, RAJENDRA ACHARYA, PRABAL BARUA, Mohammed Nafizul Islam Chowdhury Nayan
preprint en

Abstract

Abstract Background Traditional and handcrafted approaches often neglect the spatial distribution and temporal dynamics of raw EEG signals, creating a gap in objective pain classification and clinical trust. As a result, existing pain classification systems overlook important neural patterns associated with pain perception, limiting both classification performance and clinical interpretability. Objectives In this paper, we introduce a novel electroencephalography (EEG)-based approach to develop a specialised Spatial-Temporal Convolutional Neural Network (PainNet-ST) that integrates an Explainable AI (XAI) method to establish a three-class metric model for pain severity classification (low, medium, and high). Methods Our method includes the conversion of the raw EEG data from the OpenNeuro ds005280 dataset (nociceptive-evoked potentials) into multi-spectral topography maps (2D images) to preserve spatial distribution and temporal dynamics. Our proposed PainNet-ST model integrates spatial-temporal decoupling, Bayesian optimisation (implemented via the Optuna approach), and dual-layer XAI. Our analysis focused on the neural feature representations across the somatosensory cortex and the N2-P2 temporal window. Proposed PainNet-ST model was benchmarked against Support Vector Machines (SVM), Random Forest (RF), 2D-CNN, and STAM-net. For explainability, SHAP-based visualisation was incorporated to address the gap in clinical trust. Results Our proposed PainNet-ST model was evaluated against the benchmarked models with 11 performance metrics using accuracy, precision, recall, specificity, F1-score, False Positive Rate (FPR), False Negative Rate (FNR), Negative Prediction Value (NPV), False Discovery Rate (FDR), Cohen’s Kappa (k), and Area Under the Curve (AUC), with results showing that the proposed model PainNet-ST was better than the other benchmark models in every classification task. That is, in this three-class pain classification problem, an accuracy of 92.28, F1-score of 92.29%, precision of 92.35%, recall of 92.28%, and specificity of 96.14% were obtained. Conclusion In the multiclass raw EEG-based pain classification task (low, medium, high), the model reached 92.28% accuracy, demonstrating the efficacy of the proposed method for the classification of pain. The proposed PainNet-ST model successfully addresses multi-class raw EEG-based pain classification by integrating spatial-temporal decoupling, Bayesian optimisation, and dual-layer XAI.

Research Square
University of the Sunshine Coast (AU), University of Southern Queensland (AU), Fiji National University (FJ)
University of Southern Queensland
Good health and well-being
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon