Novel methodology for coronary artery disease evaluation : from a new imaging technique to deep learning based quantification

Abstract

Cardiovascular disease remains the leading cause of mortality in the world. Coronary computed tomographic angiography (CTA) has emerged as a reliable noninvasive modality for the diagnosis of coronary artery disease (CAD). However, on-site evaluation of CAD is still a challenging problem. To solve this problem, this dissertation covers methods ranging from a new imaging acquisition technique to deep learning based automatic quantification. To obtain quality CTA, catheter-directed selective CTA (S-CTA) was developed in the preclinical model, and the clinical feasibility of S-CTA was validated in patients who had diagnosed CAD. S-CTA successfully produced an optimal luminal enhancement with an extremely low-dose of iodine. Automatic quantification was developed using convolutional neural networks (CNN). We successfully measured vascular minimal lumen area, diameter stenosis, and plaque volume with the proposed CNN model. When S-CTA was used for automatic quantification, the proposed CNN successfully captured intrinsic features of the contrast-enhanced lumen and calcified plaque better than C-CTA. S-CTA can be understood as an intraprocedural CTA modality under the combined-system that incorporates the coronary angiography system and a 320-detector row CT scanner. S-CTA enables a strategic stepwise approach for coronary catheterization and on-site evaluation for coronary stenosis.