| Abstract: | Coronary Artery Disease (CAD) is the most common killer disease worldwide. To diagnosis and follow-up process of CAD is an essential requirement to control and cure the disease. Currently, Doctors perform most of the diagnoses after manual inspection of real-time coronary Computed Tomography Angiography (CTA) frame medical imaging systems. Such images investigation is tedious, time-consuming, and subject to error task for physicians. Computer-assisted CAD detection and quantification improve accuracy, save time, and minimizes human errors. Dealing with this, many pieces of research have proposed different algorithms to solve the problem. However, stenosis detection and quantification in CAD is still regarded as a challenging task. Hence, in this research, a better technique to address stenosis detection and quantification has been proposed.
In the proposed research, for detecting and quantifying stenosis (the narrow blood vessel) from the coronary artery we use CTA images. The image is first pre-processed to remove noise and enhance the contrast of the image. In the pre-processing, four alternative combinations of filtering, enhancement, and binarization have experimented; out of which image enhancement and filtering image preprocessing outperform with quality metrics parameters estimation and classification performance. Next, segmentation is used to obtain the Region of Interests (ROIs). Then, extract features using statistical and CNN based feature extraction, such as pre-trained CNN model ResNet-50 using the add-16 features layer, DenseNet-201 using fc1000 features layers, and a bag of features. The SVM and CNN classification techniques are used to identify, coronary artery stenosis presence in the CTA images. Finally, if the stenosis is present, then the stenosis quantification process to be used. The stenosis quantification is used to quantify the area of stenosis attributes. The stenosis attribute has a significant need for and importance of when determining the stent attributes to achieve the different requirements of the physiological vessels.
The performance of the proposed methods is compared with the Doctor’s opinion that can be considered a subjective score. Experimental results show that 72.7% overall accuracy achieved using an SVM classifier. The accuracy of the proposed stenosis quantification algorithm also tested and achieved 97% of overall accuracy. The mean percentage error to stenosis quantification was 1.32 from the ground truth. The result thus obtained in this study is promising to apply image processing for coronary artery stenosis detection and quantification. |
