Linear regression convolutional neural network for fully automated coronary lumen segmentation in intravascular optical coherence tomography / Yong Yan Yin

• Main Author: Yong, Yan Ling
• Format: Thesis
• Published: 2018
• Subjects: R Medicine (General); T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/9120/1/Yong_Yan_Ling.bmp; http://studentsrepo.um.edu.my/9120/11/yan_ling.pdf; http://studentsrepo.um.edu.my/9120/

Intravascular optical coherence tomography (IVOCT) is an optical imaging modality commonly used in the assessment of coronary artery diseases during percutaneous coronary intervention (PCI). Manual segmentation to assess luminal stenosis from OCT pullback scans is time consuming as each pullback contains hundreds of cross-sectional images. This segmentation is also challenging and susceptible to inter-observer variability due to various reasons including non-homogenous image intensity, blood residue, the presence and absence of different types of stents, irregular lumen shapes, image artifacts, and bifurcations. In this study, we aim to facilitate the quantitative assessment of coronary artery stenosis during PCI by developing an automatic segmentation framework to extract lumen from IVOCT images using convolutional neural network (CNN). A combination of linear-regression and convolutional neural network was proposed to automatically perform vessel lumen segmentation, parameterized in terms of radial distances from the catheter centroid in polar space. This automated segmentation algorithm has been benchmarked against manual segmentation by human experts. The proposed algorithm achieved an average locational accuracy of the vessel wall of 22 microns; 0.985 and 0.970 in Dice coefficient and Jaccard similarity index, respectively when compared against the gold standard manual segmentations. The average absolute error of luminal area estimation is 1.38 % and the processing rate is 40.6 ms per image. In addition, an inter-observer variability test was performed and has shown that the proposed algorithm has comparable variability against manual luminal area estimations by expert human observers. As a conclusion, the proposed image segmentation framework has the potential to be incorporated into a clinical workflow and to facilitate quantitative assessment of vessel lumen in an intra-operative timeframe.