Coronary artery disease (CAD) is the most common type of heart disease, killing over 370,000 Americans annu- ally2. Cardiac CT is a safe, accurate, non-invasive method widely employed for diagnosis of CAD and planning therapeutic interventions. With the current CT technology, calcium blooming artifacts severely limit the accuracy of coronary stenosis assessment. Similarly, stent blooming artifacts lead to overestimation of in-stent restenosis. As a result, many coronary CT angiography (CCTA) scans are non-diagnostic and result in patients receiving costly and invasive coronary angiography (ICA) procedures. Based on extensive feasibility results, the goal of this project is to use deep learning innovations to fundamen- tally eliminate blooming artifacts without costly redesign of the CT hardware. A consortium between GE Re- search, Rensselaer Polytechnic Institute and Weill Cornell Medicine will develop dedicated imaging protocols and machine learning methods to avoid or minimize blooming artifacts and evaluate the clinical impact of the proposed solutions.
In Aim 1, the CT scan protocol will be optimized and paired with deep learning reconstruc- tion and post-processing algorithms to generate high-resolution CT images and prevent blooming artifacts.
In Aim 2, image-domain and raw-data-domain deep learning processing algorithms will be developed to correct for residual blooming. After successful demonstration of the proposed methods on phantom scans and emulated clinical datasets, in Aim 3 the proposed CT methods will be clinically demonstrated and optimized based on 100 patients with coronary artery disease, using intravascular ultrasound as the ground-truth reference. At the end of the project, we will have demonstrated and publicly disseminated a systematic methodology to essentially remove blooming artifacts in cardiac CT without a costly hardware upgrade. This will be another suc- cess of deep learning, enabling accurate coronary stenosis assessment and eliminating many unnecessary diag- nostic catheterizations.
Blooming artifacts severely limit the accuracy of coronary stenosis assessment with cardiac CT, leading to un- necessary invasive coronary angiography procedures. The goal of this project is to eliminate blooming artifacts without costly redesign of the CT hardware, but based on optimized scan protocols and deep-learning-based image reconstruction and post-processing techniques. The proposed CT methods will be clinically demonstrated and optimized based on CT scans of 100 patients with coronary artery disease and using intravascular ultrasound as the ground-truth reference.
