ECG-based Identification of Ischemic Patterns Associated with Coronary Artery Disease Using a Hybrid Temporal Fusion Transformer-TabNet Framework on 12-lead ECG Signals
Year: 2026
Type: article
Abstract: Coronary artery disease (CAD) is still the primary cause of cardiovascular mortality in the world.Accurate automated detection systems are essential.In this paper, a novel hybrid deep learning framework is proposed whose primary novelty lies not in the TFT or TabNet architectures individually; both of them are established in their respective domains.This paper focuses on their systematic integration and joint optimisation for multimodal ECG-based identification of ischemic ECG patterns associated with CAD from 12-lead ECG signals and structured clinical metadata.The experiments were performed on the largest public annotated ECG database (PTB-XL).The PTB-XL dataset was not originally assembled as a dedicated ischemic-pattern repository.PTB-XL includes structured SCP-ECG diagnostic annotations, from which ischemic and infarction-related ECG phenotypes associated with CAD can be systematically identified.This paper provides a rigorous, clinically justified derivation of Ischemic-positive and Ischemic-negative labels from PTB-XL and then constructs a 21,837-record binary classification benchmark.The proposed framework is validated using a 5-fold stratified cross-validation method on a training set consisting of 17776 samples, with a mean validation accuracy of 0.9500.Importantly, PTB-XL's labels come from SCP-ECG diagnostic annotations, not angiography-confirmed coronary artery disease diagnoses.This study therefore focuses on identifying ischemic ECG patterns associated with CAD, rather than making a direct clinical CAD diagnosis.
Authors: Soujanya Ambala, Nirmalajyothi Narisetty
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W7235644515
