Publicação

Binary classification of cardiac pathologies using deep learning: a PTB-XL dataset approach

Ver documento

Detalhes bibliográficos
Resumo:Cardiovascular diseases, such as myocardial infarction, are among the leading causes of death worldwide. Accuracy and time are crucial for diagnosing these conditions and for effective treatment, usually requiring time-consuming manual analysis of clinical-grade electrocardiogram (ECG). This paper presents a novel deep learning-based method for binary classification of cardiac patholo-gies using the PTB-XL dataset. The model integrates EfficientNetB3 for spatial feature extraction and a Linformer block to capture long-range dependencies be-tween leads. Preprocessing involves converting RGBA ECG images to RGB for-mat and normalizing them to meet the requirements of the inputs of the layers. Initial experiments have shown promising results, achieving an AUC (Area Un-der the Curve) of 86.06%. Further work includes tests to optimize the model's performance based on different key metrics, including accuracy and precision.
Autores principais:Chaabani, Mohamed
Outros Autores:Guerreiro, Nathan; Ribeiro, Luiz; Luiz, Luiz E.; Slim, Mohamed; Teixeira, João Paulo
Ano:2025
País:Portugal
Tipo de documento:documento de conferência
Tipo de acesso:acesso aberto
Instituição associada:Instituto Politécnico de Bragança
Idioma:inglês
Origem:Biblioteca Digital do IPB
Descrição
Resumo:Cardiovascular diseases, such as myocardial infarction, are among the leading causes of death worldwide. Accuracy and time are crucial for diagnosing these conditions and for effective treatment, usually requiring time-consuming manual analysis of clinical-grade electrocardiogram (ECG). This paper presents a novel deep learning-based method for binary classification of cardiac patholo-gies using the PTB-XL dataset. The model integrates EfficientNetB3 for spatial feature extraction and a Linformer block to capture long-range dependencies be-tween leads. Preprocessing involves converting RGBA ECG images to RGB for-mat and normalizing them to meet the requirements of the inputs of the layers. Initial experiments have shown promising results, achieving an AUC (Area Un-der the Curve) of 86.06%. Further work includes tests to optimize the model's performance based on different key metrics, including accuracy and precision.

Atividades financiadas

Carregando projetos financiados...