Analysis of Preprocessing Data Approaches in Stenosis Detection Based on X-Ray Coronary Angiography Images Using Faster R-CNN

Abstract

Coronary Artery Disease (CAD) is a form of cardiovascular disease that is the leading cause of death worldwide. CAD occurs due to plaque buildup, which consists mainly of cholesterol and cell metabolism waste on the artery walls. This buildup causes stenosis, which is an abnormal narrowing of the coronary arteries that obstructs blood flow. One of the gold standard examination methods for detecting stenosis is X-ray coronary angiography. This study aims to analyze the effect of preprocessing stages on the performance of the Faster R-CNN model in detecting stenosis in X-ray coronary angiography images. Two training scenarios were conducted, namely a model without preprocessing and a model with preprocessing stages that included black border removal, denoising, and contrast enhancement using CLAHE. The dataset was divided with a ratio of 80:20 for training and testing data, and the model was trained for 10 epochs using the ResNet-50 FPN backbone. The evaluation results show that the model with preprocessing performs better than the model without preprocessing, with a precision value of 0.8050, recall of 0.9718, F1-Score of 0.8805, and Mean IoU of 0.7596.