Avaliação de redes neurais no reconhecimento de complexo QRS em sinais de eletrocardiografia

Abstract

Electrocardiographic recordings are considered to be a highly important tool in the diagnosis of cardiac disease. So as to make a proper evaluation of the heart condition, one must accurately detect the QRS-complex during the analysis of the electrocardiographic (ECG) signal, often corrupted by noise. Such fact hinders the routines of the professionals who analyze this type of medical recording. In consequence, tools that improve the QRS-complex detection are required. The aim of this project is the development of a computing system for QRS-complex identification in ECG signals. The methodology employed in this project comprises the stages of project, development and tests. The development and tests stages have used an ECG database available in the public domain. The tests stage was performed by the team responsible for this project by using a tool built in the system for this specific purpose. The system is composed of the following modules: ECG signal visualizing, event marking, characteristic extraction, pattern classification and results evaluation. The characteristic extraction module correlates a pattern and the ECG signal, producing a new signal, which works as input for the pattern classification module. The events detected by this pattern classification module are graphically exhibited on the computer screen. Finally, the evaluation module compares the results obtained from the system with the annotations from the database, and classify them as true positive, false positive, true negative or false negative. Based on this evaluation, both sensibility and specificity analysis were calculated, which allows the plot of a ROC graph. The achievements of this project, therefore, consist of the system itself and its practical usage.