Deteção de enfarte do miocárdio através de redes neuronais convolucionais

Introduction: Cardiovascular diseases have a high mortality rate worldwide. The Electrocardiogram is the first-line exam in what concerns the diagnosis of this type of pathologies and, consequently, with extreme importance in the correct and immediate interpretation for a promising prognosis. Myocardial infarction is one of the electrocardiographic changes that detected in a timely manner can have a huge impact at the physiological and anatomical level of the cardiac muscle itself. The existing need for a fast and correct decision-making has led to the development of algorithms capable of detecting pathologies in the electrocardiographic signal. Methodology: In order to maximize the discriminative capacity of the different types of myocardial infarction, specific patterns have been extracted from the ECG signals to feed artificial intelligence algorithms. In order to make the best use of the artificial intelligence algorithms, a pre-processing of the entire signal was performed followed by a rigorous selection of the segments that show pathological activity for each disease. The pathological segment selection for feeding the CNN was made by comparing the segments over time with time-series-sequency model’s characteristic of myocardial infarction. Results: The precision values of the models used in the present study, presented accuracy levels above 97%, 99.39%, 99.64%, 97.76% and 98.98% for the Anterior, Anterolateral, Inferior and Inferolateral, respectively. The promising discriminative results prove that the segment model selection stage allow to perform an excellent screening of the pathological sequencies from the nonpathological time-series sequences and the CNN model is ready for detecting pathological activity over time as a new ECG signal is presented to its entries.