An Ensemble Model of DL for ECG-Based Human Identification

• Published in: IEEE transactions on instrumentation and measurement Vol. 73; pp. 1 - 15
• Main Authors: Begum, Ritu Nazneen Ara, Sharma, Ambalika, Singh, Girish Kumar
• Format: Journal Article
• Language: English
• Published: New York IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Arrhythmia; Artificial neural networks; Biological system modeling; Biometrics (access control); Datasets; Electrocardiogram (ECG)-based human identification; Electrocardiography; Feature extraction; heart rate variability (HRV); intra-subject variability; Recording; Rejection rate; Sensitivity; Tachycardia; U-Net
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2024.3385842

In this study, an ensemble model of U-Net and artificial neural network (ANN) is developed for electrocardiogram (ECG)-based human identification. This study specifically presents the impact of the intra-subject variability parameter on the efficacy of the model. Extensive experimentation using a dataset of ECG recordings of 620 users from 12 different databases is carried out. The model is tested on each of the 12 datasets separately as well. Metrics like recall, precision, <inline-formula> <tex-math notation="LaTeX">F1 </tex-math></inline-formula>-score, sensitivity, specificity, false acceptance rate (FAR), and false rejection rate (FRR) have been used to evaluate the model. The overall identification accuracy of the proposed ensemble model is found to be 98.21% for 620 users. For the individual dataset, the model scored highest with the Medical Information Mart for Intensive Care (MIMIC) II/III dataset with an identification rate (IDR) of 99.84% and lowest with ventricular tachycardia beat (VTB) and Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) dataset with an IDR of 94.00% each. The highest sensitivity reached is 0.998 for MIMIC II/III and the highest specificity obtained is 0.999 for the intracardiac atrial fibrillation (IAF) dataset. The results obtained show that the proposed ensemble network is a promising model for ECG-based human identification.