Classificação automática da qualidade de sinal de ECG com redes neurais convolucionais

• Autor(a) principal: Marinho, Ayrton Sousa
• Data de Publicação: 2021
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Federal do Ceará (UFC)
• Texto Completo: http://www.repositorio.ufc.br/handle/riufc/64333
• topic: Eletrocardiograma; Deep Learning; Redes Neurais Convolucionais; Machine Learning

According to World Health Organization (WHO) Organization (2017), many deaths worldwide occur due to heart problems and tools, such as, Electrocardiogram (EKG) are used to try to identify possible heart problems. An automated solution has been tested through several studies in which they seek to predict, through an intelligent analysis of a EKG signal, if an individual has any heart disease. However, some EKG signals are captured using devices that are often not efficient in delivering the signal quality that is commonly expressed by noise or inconsistencies in the signal. This paper presents as a central theme the quality classification of EKG signals using Deep Learning techniques to learn the pre-existing signals and classify them in three possible categories: Acceptable, Semi-Acceptable and Not Acceptable. Thus, the present paper seeks to verify the feasibility of using an automatic solution using DL techniques with a focus on CNN to classify the quality of these signals, in order to ensure better filtering and guarantee quality of data for training in disease prediction models. The methodology, it is an exploratory research with a qualitative nature, a database already created specifically for the task of classification of EKG signals was used. As a metric for the classification of the algorithm, the F-Score, Precision and Recall metrics obtained through the random search technique in 100 different samples of models are used. The results were positive and indicate an F-SCORE of 0.9270, Precision of 0.9278 and a Recall of 0.9264 for the best combination of hyperparameters based on the data used for the training, testing and validation of the models.