Aquisição de eletrocardiograma e classificação de arritmias por machine learning
Carregando...
Citações na Scopus
Tipo de produção
Trabalho de Conclusão de Curso
Data
2022-08-30
Autores
Silva, César Nogueira
Matos, Jefferson Almeida
Lopes, Fernanda Ferrezi
Matos, Jefferson Almeida
Lopes, Fernanda Ferrezi
Orientador
Castro, Maria Claudia F.
Periódico
Título da Revista
ISSN da Revista
Título de Volume
Citação
Texto completo (DOI)
Palavras-chave
ECG,machine learning,arritmia,MATLAB,arrhythmia
Resumo
Esse projeto apresenta uma proposta de um dispositivo para monitoração cardíaca
pessoal. O dispositivo foi desenvolvido aplicando técnicas de Inteligência Artificial
(aprendizado de máquina) para o reconhecimento de arritmias. O dispositivo informará, por meio de um e-mail, a detecção da possível arritmia à um médico ou pessoa credenciada, de modo que possa agilizar ou antecipar, um atendimento de emergência. Segundo a classificação da AAMI (Association for the Advancement of Medical Instrumentation) foram selecionadas 4 classes, sendo Normal (N), Supraventricular ectópico (SVE), Ventricular ectópico (VE) e Fusão de normal e ventriculares (F). Utilizando a base de dados MIT-BIH Arrhythmia Database e o
aplicativo (app) Classification Learner do Matlab, para o treinamento, foi possível investigar vários modelos, sendo que os melhores foram o Ensemble (SubspaceKNN) e o SVM (Cubic SVM) com acurácias de 94% e 94,1%, respectivamente. Esses modelos foram selecionados para a etapa de teste, tendo o modelo Ensemble (SubspaceKNN) obtido a melhor acurácia (74,4%) e posteriormente sendo utilizado para a implementação no aplicativo de interface do
usuário. Desta forma, sua implantação contribui para o desenvolvimento de inovação científicotecnológica, à medida que o equipamento apresenta características que ainda não foram totalmente implementadas no mercado e causa impacto no âmbito social, tornando-o acessível, sem a necessidade de um serviço de monitoração contínuo contratado pelo paciente.
This project presents a proposal for a device for personal cardiac monitoring. The device was developed by applying Artificial Intelligence (machine learning) techniques for the recognition of arrhythmias. The device will inform, by means of an e-mail, the detection of the possible arrhythmia to a doctor or accredited person, so that he can speed up or anticipate an emergency care. According to the AAMI (Association for the Advancement of Medical Instrumentation) classification, 4 classes were selected, being Normal (N), Supraventricular ectopic (SVE), Ventricular ectopic (LV) and Fusion of normal and ventricular (F). Using the MIT-BIH Arrhythmia Database and the application (app) Classification Learner from Matlab, for training, it was possible to investigate several models, the best of which were the Ensemble (SubspaceKNN) and the SVM (Cubic SVM) with accuracies of 94% and 94.1%, respectively. These models were selected for the test stage, and the Ensemble model (SubspaceKNN) obtained the best accuracy (74.4%) and was later used for implementation in the user interface application. In this way, its implementation contributes to the development of scientific-technological innovation, as the equipment presents characteristics that have not yet been fully implemented in the market and causes an impact on the social sphere, making it accessible, without the need for a monitoring service. continuous contracted by the patient.
This project presents a proposal for a device for personal cardiac monitoring. The device was developed by applying Artificial Intelligence (machine learning) techniques for the recognition of arrhythmias. The device will inform, by means of an e-mail, the detection of the possible arrhythmia to a doctor or accredited person, so that he can speed up or anticipate an emergency care. According to the AAMI (Association for the Advancement of Medical Instrumentation) classification, 4 classes were selected, being Normal (N), Supraventricular ectopic (SVE), Ventricular ectopic (LV) and Fusion of normal and ventricular (F). Using the MIT-BIH Arrhythmia Database and the application (app) Classification Learner from Matlab, for training, it was possible to investigate several models, the best of which were the Ensemble (SubspaceKNN) and the SVM (Cubic SVM) with accuracies of 94% and 94.1%, respectively. These models were selected for the test stage, and the Ensemble model (SubspaceKNN) obtained the best accuracy (74.4%) and was later used for implementation in the user interface application. In this way, its implementation contributes to the development of scientific-technological innovation, as the equipment presents characteristics that have not yet been fully implemented in the market and causes an impact on the social sphere, making it accessible, without the need for a monitoring service. continuous contracted by the patient.