Greedy Soft Matching for Vascular Tracking of Coronary Angiographic Image Sequences

Abstract

Purpose: Vascular tracking of angiographic image sequences is one of the most clinically important tasks in the diagnostic assessment and interventional guidance of cardiac disease due to providing dynamic structural information for precise motion analysis and 3D+t reconstruction. However, this task can be challenging to accomplish because of unsatisfactory angiography image quality and complex vascular structures. Thus, this study converted vascular tracking into branch matching and proposed a new greedy graph search-based method for it.Methods: Each vascular branch was separated from the vasculature and was tracked independently.Then, all branches were combined using topology optimization, thereby resulting in complete vasculature tracking. An intensity-based image registration method was applied to determine the tracking range, and the deformation field between two consecutive frames was calculated. The vascular branch was described using a vascular centerline extraction method with multi-probability fusion-based topology optimization. We introduced an undirected acyclic graph establishment technique. A greedy search method was proposed to acquire all possible paths in the graph that might match the tracked vascular branch. The final tracking result was selected by branch matching using dynamic time warping with a DAISY descriptor.Results: For single branch dataset SBD, the proposed method was evaluated on 12 angiographic image sequences with 77 angiograms of contrast agent-filled vessels. The average precision, sensitivity and F1 score of the tracking result of all angiograms were 0.90, 0.89 and 0.89, respectively. The average F1 score of the tracking results of the first, middle and last frames in all sequences were 0.91, 0.91 and 0.86, respectively. In the vessel tree dataset VTD, the proposed method was validated on 9 angiographic image sequences with 58 angiograms of contrast agentfilled vessels. The average precision, sensitivity and F1 score of the tracking result of all angiograms were 0.89, 0.87 and 0.88, respectively. The average F1 score of the tracking results of the first, middle and last frames in all sequences were 0.94, 0.88 and 0.82, respectively. Compared with five other state-of-the-art methods, the proposed method accurately tracked the vasculature from angiographic image sequences and the results were insignificantly affected by the tracking span in both datasets. Conclusions:The solution to the problem reflected both the spatial and textural information between successive frames. Thus, the proposed method is robust and highly effective in vascular tracking of angiographic image sequences and the approach provided a universal solution to address the problem of filamentary structure tracking.

Real-time coronary artery stenosis detection based on modern neural networks

“…These approaches follow the general scheme: (1) coronary artery tree extraction, (2) calculation of geometric dimensioning, and (3) analysis of the stenotic segment. The key stage that determines the speed and accuracy of such algorithms is based on the coronary artery tree extraction using the centerline extraction 8 , 9 ; the graph-based method 10 – 12 ; superpixel mapping 13 , 14 ; and machine/deep learning 15 – 17 . The last, being a powerful tool for computer vision and image classification, has shown great promise in CAD detection due to their performance, tuning flexibility, and optimization.…”

“…Albeit several CNN-based approaches focused on achieving optimal accuracy for CAD detection with the Dice Similarity Coefficient of more than 0.75 12 , 13 and/or the Sensitivity metric of more than 0.70 21 have been proposed, their speed remains disregarded. Image processing time is an important indicator for the applied use of these methods that can reach 1.1–11.87 s 10 , 20 s 10 , 13 , and over 60 s 9 . However, this time is unacceptable for real-time CAD detection with the processing rate of 7.5–15 fps instead of the required 0.13–0.07 s per frame 22 , 23 .…”

Open Curvature Scale Space Matching for Coronary Artery Identification in X-Ray Angiographic Images

“…The three-dimensional reconstruction of coronary arteries can provide information on the coronary artery spatial structure, quantitative analysis of the ratio of blood vessel coverage and projection shortening [10]- [12], information on stenotic lesions [13]- [16], etc. In addition, the identification and matching of coronary structures are the foundation for coronary dynamic remodeling [17]- [19] and hemodynamic calculations [20]- [21] in the cardiac cycle.…”

Real-Time Coronary Artery Stenosis Detection Based on Modern Neural Networks

“…Recent studies commonly use the Dice Similarity Coe cient (some studies reported the Dice Similarity Coe cient of more than 0.75) 11,12 and/or the Sensitivity metric (some studies reported the Sensitivity of more than 0.70) 17 to assess the quality of automatic CAD analysis. Image processing time is an important indicator for the applied use of these methods that can reach 1.1-11.87 seconds 9 , 20 seconds 9,12 , and over 60 seconds 8 . Taking into account the mean duration of angiography imaging series, usually consisting of 50-80 frames, the total processing time can become a signi cant factor, limiting the use of many methods.…”

“…The key stage that determines the speed and accuracy of such algorithms is based on the coronary artery tree extraction using the centerline extraction 7,8 ; the graph-based method [9][10][11] ; superpixel mapping 12,13 ; and machine/deep learning [14][15][16] .…”