Disease-Specific Reduced-Lead ECG Classification Using EMD and Firefly Selection
Disease-Specific Reduced-Lead ECG Classification Using EMD and Firefly Selection
Amir hossein Abbasi1 Azam Sheikhian2 Hadi Pazhakhzadeh3
1) M.Sc. Biomedical Engineering, Isfahan University of Medical Sciences,
2) Ph.D. Computer Software Engineering, Islamic Azad university, Yasuj Branch,
3) Ph.D. Technology Management, Islamic Azad university, Kish Island Branch,
IOI Code
:
AAHD-ABHEAK
Publication :
National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
Electrocardiogram (ECG) analysis is a commonly used non-invasive method for evaluating cardiac abnormalities. While the conventional 12-lead ECG provides valuable diagnostic information, the use of all leads increases the input dimensionality and computational cost of automated classification systems. Recent studies have investigated reduced-lead ECG analysis; however, many existing approaches rely on predefined lead configurations and overlook the fact that the diagnostic relevance of ECG leads can differ depending on the cardiac condition. In this study, a disease-specific reduced-lead ECG classification framework is proposed by integrating Empirical Mode Decomposition (EMD) and Firefly-based lead selection. The PTB-XL database was used, and five diagnostic superclasses, including normal ECG (NORM), myocardial infarction (MI), ST/T changes (STTC), conduction disturbance (CD), and hypertrophy (HYP), were considered. Following signal preprocessing, EMD-based features consisting of statistical, temporal, energy, entropy, and Hilbert transform descriptors were extracted from individual ECG leads. The Firefly algorithm was then employed to determine the most informative lead subsets for each disease-versus-normal classification task. Four binary classification scenarios, including MI vs NORM, STTC vs NORM, CD vs NORM, and HYP vs NORM, were evaluated using Gradient Boosting, Lead-Attentive Neural Network, and Support Vector Machine (SVM) classifiers. Evaluation on the independent test set showed that the selected lead subsets varied across diagnostic classes. SVM achieved the highest AUC values for MI, STTC, and CD classification with 91.64%, 91.84%, and 89.82%, respectively, whereas Gradient Boosting provided the best results for HYP classification, achieving an accuracy of 93.73% and an AUC of 94.00%. These results demonstrate that Firefly-based lead selection combined with EMD-derived features can reduce the reliance on multiple ECG leads while preserving reliable classification performance.
Keywords : ecg classification reduced lead ecg firefly optimization emd