Deep Learning-Based Approach to Automatically Assess Coronary Distensibility Following Kawasaki Disease

Kawasaki disease is an acute vasculitis affecting children, which can lead to coronary artery (CA) aneurysms. Optical coherence tomography (OCT) has identified CA wall damage in KD patients, but it is unclear if these findings correlate with any distensibility changes in the CA and how these changes evolve over time. This paper seeks to establish the link between OCT findings and vessel distensibility with a novel deep learning coronary artery segmentation system and use the segmentation framework to automatically analyze the temporal evolution of coronary stiffness over many years. 27 KD patients underwent catheterization with coronary angiography of the left coronary artery (LCA), followed by OCT of proximal and distal segments of the LCA. Changes in the CA caliber over the cardiac cycle were measured automatically and compared against OCT findings suggestive of KD-related vascular damage. In addition, 34 KD patients with regressed or persistent CA aneurysms were followed with serial CA angiography over an average of 14.5 years. Distensibility changes were calculated using a deep learning coronary artery segmentation framework and evaluated longitudinally. Distensibility in the coronary arteries after KD negatively correlated with increasing severity of OCT findings of KD-related vessel damage. KD patients have a significant increase in CA wall stiffness at 1 year after diagnosis, which then plateaus subsequently, compared to controls. Also, patients with persistent CA aneurysms have a statistically significant increase in wall stiffness over time in comparison to those with regressed CA aneurysms. Distensibility changes in the CA of KD patients calculated using our automated deep learning approach correlates with the severity of OCT findings of KD-related CA damage. This decreased distensibility peaks at 1 year in KD patients when following longitudinally and is more severe in those with persistent CA aneurysms.