Título: Deep Learning Models for Real -World Detection of Electrode Interchange in 12 – Lead ECGs: A Telecardiology Validation
Autores: Isac Andrade Alves, Gabriela Miana, Paulo Rodrigues Gomes, Jessyca A. Bessa, Antônio Luiz P. Ribeiro & Jermana Lopes de Moraes
Resumo: Electrode interchanges during 12-lead electrocardiogram (ECG) signal acquisition is a recurring problem, capable of compromising diagnostic accuracy and generating additional costs for both the patient and the healthcare system. The automatic detection of these interchanges is essential to support clinical decision-making and improve outcomes in cardiovascular care. In this work, different combinations of deep learning architectures are investigated for the automatic detection of clinically relevant electrode interchanges, RA-LA, RA-LL, V1-V6, V2-V5 and complete precordial electrode interchanges. The models explored in this work are Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Bidirectional Long Short-Term Memory (BiLSTM) and Bidirectional Gated Recurrent Unit (BiGRU). The training and evaluation of the models were performed with a database composed of 22,886 ECGs, unbalanced and representative of a real telecardiology scenario in Brazil. In general, the model based on the CNN+RNN presented consistent performance in all classes. For the ECG without interchange and RA-LA, RA-LL, the model achieved metrics with values greater than or to 0.96, while for inversions in the precordial electrodes, the metrics were greater than 0.86. Furthermore, the CNN+RNN model exhibited the lowest number of parameters and memory usage among the evaluated architectures, highlighting its computational efficiency. The results demonstrate the feasibility and effectiveness of using deep learning techniques in the automated detection of electrode interchanges. The incorporation of the proposed models in cardiology services brings direct benefits to the entire workflow of the health system.
Palavras-chave: Electrode interchange; 12-lead ECG; Deep learning; real-world clinical data; imbalanced dataset; telecardiology.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1191460
Artigo em PDF: CBIC_2025_paper1191460.pdf
Arquivo BibTeX:
CBIC_2025_1191460.bib