Detection of coronary artery blockage at an early stage using effective deep learning technique

Abstract

A coronary artery blockage is a form of coronary artery disease also known as CAD. It is the most common and frequent disease affecting the human body over the age of 65. CAD is a type of cardiovascular disease that happens because of a disorder in the coronary arteries of the human heart. Stenosis is the abnormal narrowing of the coronary artery due to the buildup of cholesterol which results in poor blood circulation causing a blockage. The development of computer science technologies has made drastic changes in medical science practices that include cardiology. Such

advancements have made the invention of medical tests like Angiogram, Electrocardiogram, Magnetic Resonance Imaging, Echocardiogram, etc. These are imaging

techniques to visualize arterial and venous vessels throughout the body for the diagnosis of various diseases. In common medical practice, the analysis and diagnosis

of CAD mainly rely on the visual inspection and calculation of measurements by experienced cardiologists and doctors. Our proposed method aims toward a fully automated system for detecting a coronary artery blockage at an early stage using image processing and deep learning techniques so that the system can help doctors as well as patients to improve the medical treatment of the heart at an early stage. The goal of this research is to implement a system that can detect stenosis areas of the coronary artery due to the buildup of cholesterol plaque and other blocking agents. To examine stenosis in the coronary artery, Angiogram images are essential. Evaluating 2,151 Angiogram Image Dataset we train and test our models to reach a conclusion. The research uses CNN architecture models that use a dataset of 2D Angiogram images of the segmented coronary arteries which are analyzed using VGG16, VGG19, ResNet50, InceptionV3, and DenseNet121 models. To enhance our study, we classified our dataset into two classes i.e Binary Classification and Multiclass Classification. Next, using Ensemble model architecture, we evaluate the results and accuracy of the models used in the procedure of identifying coronary artery blockage. We used evaluation metrics Accuracy, Precision, Recall, and F1 Score to evaluate our results. Finally, we achieved accuracy, precision, recall, and F1 Score of more than 0.99 for the Binary Classification and more than 0.98 for the multiclass classification respectively of our dataset. In this way, the use of deep learning techniques can improve and develop medical science at a prodigious level resulting in error-free medical treatment of the heart at an early stage.