Research On Co-registration Between Coronary Angiography And OCT Based On Lumen And Branch Segmentation

Coronary angiography provides real-time and clear imaging of the coronary anatomy.To a certain extent,OCT makes up for the deficiency of coronary angiography,because it allows detailed assessment of luminal dimension and plaque morphology.Automatic co-registration of coronary angiography and OCT images,which can reinforce complementary advantages of the two imaging technologies,is of great significance for clinical treatment.Aiming at the shortcomings of the current OCT-angiography co-registration methods,a new method for automatic OCTangiography co-registration based on lumen and branch segmentation was proposed.The main work is summarized as follows:(1)According to the characteristics of coronary angiography and OCT images,an overall scheme of OCT-angiography co-registration based on lumen and branch segmentation was proposed.Co-registration scheme began with the detection and segmentation of the main vessel and branch ostia in coronary angiography and OCT images,respectively.Then lumen feature point sets and branch feature point sets of coronary angiography and OCT images were constructed respectively based on lumen and branch segmentation.The feature point sets were registrated by a two-stage registration method including fast target detection with sliding window in coarse registration and TPS-RPM-LDB in precise registration,thereby the corresponding relation of angiography and OCT images were obtained.(2)According to the characteristics of coronary angiography images,the CABMask-RCNN model was proposed based on Mask-RCNN model for the detection and segmentation of the branch ostia on the main vessel in coronary angiography images.In order to correctly distinguish the main and non-main vessels,both original coronary angiography images and mask images of the main vessel were used as multichannel inputs to provide enhanced vascular structure information.At the same time,the attention mechanism was integrated into the feature extraction network and improve the feature extraction ability of the network.Compared with the original mask-RCNN model,the CAB-mask-RCNN model improved 0.2853 in target detection and 0.3150 in segmentation.(3)The feature point sets for the two-stage registration between angiography and OCT images were constructed respectively based on lumen and branch segmentation.In the precise registration,the TPS-RPM-LDB algorithm,which made full use of the characteristics of coronary vessel,was proposed for more accurate registration.It added local diameter information constraint and branch topology preservation constraint.The two-step co-registration algorithm was validated in the clinical data set.The mean co-registration error at the marker level was 0.38 mm,and the co-registration accuracy at the marker level was 95.82%.The algorithm proposed in this paper has high accuracy and strong robustness in different coronary vessels.(4)A clinical data set was used to validate the coronary angiography and OCT image co-registration algorithm with a single vessel as the research target.A total of212 vessels from 208 patients were analyzed.Co-registration was achieved in 208 out of 212 vessels,with an accuracy of 98.11%.The novel method in this paper for automatic OCT-angiography co-registration is feasible and accurate without altering the basic diagnostic workflow in the cathlab with additional acquisitions,so it can be implemented in routine clinical practice.