ECG classification using Artificial Intelligence: Model Optimization and Robustness Assessment

The Electrocardiogram is one of the more complete exams for diagnosing pathologies regarding the cardiovascular system. Therefore, and based on the rudimentary methods of analyzing these exams in the past, computer-based approaches are now used in ECG analysis. Smart technology systems have been designed over time to diagnose cardiovascular conditions through ECG analysis. The current study evaluates the robustness and performance of one Artificial Intelligent (AI) system, based on a Convolutional Neural Network paired with a Multilayer Perceptron, to classify specific cardiac conditions in ECG signals. The paper assesses the robustness of the described AI model, by evaluating its performance in the classification of different classes. Moreover, it was studied the influence of the model's parameters in the result, namely: Train-Test split ratio, learning rate, optimization Algorithm, and a number of epochs. After finding the optimal parameterization configuration that translates into a higher and better accuracy of the system, the results suggested that heartbeat classification based on CNN+MLP architecture is robust and capable to deal with the class increase. Our goal is to use outcomes and intelligent systems to automatically process ECG signals, and to directly identify specific medical conditions, and, for that, we intend to study the influence of the parameter's variation and confirm its robustness to the variation of the database configuration.