Study On The Automated Analysis Of Bioresorbable Vascular Scaffold In Intravascular Optical Coherence Tomography Images

Coronary artery disease(CAD)is currently the most deadly disease in the world.Percutaneous coronary intervention(PCI)is a common treatment modality for CAD,which relieves the narrowing of coronary artery by coronary catheterization and keeps the blood vessel open by stent implantation,so that blood supply is improved and the risk of CAD is decreased.Currently,the latest stent type is the bioresorbable vascular scaffold(BVS)which is made of bioabsorbable polymeric materials.BVS is seen as the most promising type of stent in the future.Intravascular optical coherence tomography(IVOCT),due to its superior resolution,has become one of the most frequently-used intracavitary imaging techniques in interventional cardiology.IVOCT offers excellent imaging quality for internal structures of blood vessel,and can assist cardiologists to conduct stent struts analysis during stent implantation,which is of important value in theoretical research and practical application.However,since an IVOCT pullback contains as many as hundreds of images and thousands of struts,it is time consuming and labor intensive to manually analyze IVOCT images,and can hardly meet the clinical requirement of real time.This paper proposes an automated method for BVS struts analysis in IVOCT images.We utilize Adaboost algorithm to train cascade classifiers for the detection of struts positions and sizes,and then employ deep learning to further improve the detection accuracy.Based on the detection results,we transform each strut into polar coordinate system by Hough transformation and search for the optimal path by dynamic programming,so that struts boundaries are automatically segmented.Based on the segmentation results,struts malapposition analysis is automatically conducted to assist cardiologists to make diagnoses in the clinical operation.In order to evaluate the comprehensive performance of our method,testing experiment was conducted on 5 IVOCT pullbacks in which we compared the detection and segmentation results between our method and the ground truth labeled by experts.Experimental results showed that the proposed Adaboost-based detection reached a recall rate of 91.3% and precision of 89.6% on average,and the deep learning-based detection further improved the recall rate and precision to 97.9% and 95.2% respectively.The average Dice coefficient was 0.812 for struts boundary segmentation.For a single pullback,the time consumption of Adaboost-based detection and R-FCN-based detection was 30.21 seconds and 14.64 seconds respectively,and the segmentation cost 32.34 seconds.It suggested that our study can realize automated detection and segmentation for BVS struts in IVOCT images,and meet the clinical requirement of real time.