Application of spatial uncertainty predictor in CNN-BiLSTM model using coronary artery disease ECG signals
Metadatos
Mostrar el registro completo del ítemAutor
Seoni, Silvia; Molinari, Filippo; Acharya, U. Rajendra; Shu Lih, Oh; Datta Barua, Prabal; García López, Salvador; Salvi, MassimoEditorial
Elsevier
Materia
Explainable AI CAD ECG
Fecha
2024-03-02Referencia bibliográfica
Seoni, Silvia, et al. Application of spatial uncertainty predictor in CNN-BiLSTM model using coronary artery disease ECG signals. Information Sciences 665 (2024) 120383 [10.1016/j.ins.2024.120383]
Resumen
This study aims to address the need for reliable diagnosis of coronary artery disease (CAD) using
artificial intelligence (AI) models. Despite the progress made in mitigating opacity with
explainable AI (XAI) and uncertainty quantification (UQ), understanding the real-world predictive
reliability of AI methods remains a challenge. In this study, we propose a novel indicator
called the Spatial Uncertainty Estimator (SUE) to assess the prediction reliability of classification
networks in practical Electrocardiography (ECG) scenarios. SUE quantifies the spatial overlap of
critical Grad-CAM (Gradient-weighted Class Activation Mapping) features, offering a confidence
score for predictions.
To validate SUE, we designed a deep learning network that integrates Convolutional Neural
Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) mechanisms for precise
ECG signal classification of CAD. This network achieved high accuracy, sensitivity, and specificity
rates of 99.6%, 99.8%, and 98.2%, respectively. During test time, SUE accurately distinguishes
between correctly classified and misclassified ECG segments, demonstrating the superiority of the
proposed network over existing methods.
The study highlights the potential of combining XAI and UQ techniques to enhance ECG
analysis. The evaluation of spatial overlap among discriminative features provides quantitative
insights into the network’s robustness, encompassing both current prediction accuracy and the
repeatability of predictions.