Título: Aplicação de uma Rede Neural Feedforward com Algoritmo de Levenberg-Marquardt para Classificação de Alterações do Segmento ST do Eletrocardiograma
Autores: Soares, Pedro Paulo da Silva; Nadal, Jurandir
Resumo: A feedforward artificial neural network (ANN) trained with the Levenberg-Marquardt algorithm was employed to classify electrocardiographic ST segments into three classes: normal (N), negative (ST-) and positive ST deviations (ST+). Each segment was resampled to 100 points and reduced by principal component analysis (PCA) to five coefficients, which were the inputs of the ANN with fifteen neurons in the hidden layer and three outputs. A total of 7200 patterns were randomly selected, being 2400 of each class, and equally divided to create the training (TRE), validation (VAL) and test (TES) sets. TRE was used for PCA calculation, where the five principal components represented 98.23% of the original data variance. TRE was used for training the ANN, and VAL for validation, where the mean square error increasing over five consecutive epochs was the criterion to stop the training. The measured performance for TES data set reached 80.42% accuracy, 84.86% sensitivity and 87.82 % positive predictivity, which are similar to other methods in literature. The conclusion is that the method using PCA and ANN is potentially applicable for the electrocardiographic ST changes classification.
Artigo em pdf: 4cbrn_092.pdf
Arquivo BibTex: 4cbrn_092.bib