A Survey on Automated Detection of Coronary Artery Disease in CT Angiography Using Recurrent CNN Approaches

Abstract

Coronary artery disease (CAD) is a significant global cause of disability and mortality, underscoring the need for advanced imaging techniques to detect and classify early indications of artery plaque and stenosis.By integrating Recurrent Convolutional Neural Networks (R-CNN) with transfer learning, this study introduces a new method that may significantly enhance the precision and consistency of coronary CT angiography (CCTA) diagnoses of serious cardiovascular diseases.The proposed R-CNN model synergistically integrates the capabilities of convolutional layers for extracting spatial features with recurrent layers for learning temporal sequences. This integration enhances the ability to identify complex patterns related to CAD. Our dataset consists of a diverse collection of CCTA scans with annotated regions of interest, covering normal coronary anatomy, various plaque types, and different degrees of stenosis. The R-CNN is trained to automatically detect and classify these regions, offering a comprehensive assessment of coronary vasculature health. The recurrent nature of the network allows it to capture temporal information, which is crucial for accurately characterizing dynamic changes in plaque composition and stenosis severity. Experimental results demonstrate an accuracy rate of more than 99%, showing the accuracy of our method. Thanks to its ability to learn from a pre-trained model and its capacity to acquire complex hierarchical features from the input, the model achieves remarkable accuracy.The robust performance of the R-CNN is further validated through extensive testing on an independent dataset, showcasing its potential for clinical application in real-world scenarios.