Automatic 3D adaptive vessel segmentation based on linear relationship between intensity and complex-decorrelation in optical coherence tomography angiography

Yiming Zhang; Huakun Li; Tongtong Cao; Ruixiang Chen; Haixia Qiu; Ying Gu; Peng Li

doi:10.21037/qims-20-868

Automatic 3D adaptive vessel segmentation based on linear relationship between intensity and complex-decorrelation in optical coherence tomography angiography

Quant Imaging Med Surg. 2021 Mar;11(3):895-906. doi: 10.21037/qims-20-868.

Authors

Yiming Zhang¹, Huakun Li¹, Tongtong Cao¹, Ruixiang Chen¹, Haixia Qiu², Ying Gu², Peng Li¹

Affiliations

¹ State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, International research center for advanced photonics, Zhejiang University, Hangzhou, China.
² Department of Laser Medicine, First Medical Center of PLA General Hospital, Beijing, China.

Abstract

Background: Vascular quantitative metrics have been widely used in the preclinical studies and clinical applications (e.g., the diagnosis and treatment of port wine stain, PWS), which require accurate vessel segmentation. An automatic 3D adaptive vessel segmentation is in need for a reproducible and objective quantification of the optical coherence tomography angiography (OCTA) image.

Methods: Human skin imaging was performed with a lab-built optical coherence tomography (OCT) system. Rather than separately applying the conventional 2-step (intensity and binarization) thresholding in the decorrelation-contrast OCTA, we proposed a 3D adaptive threshold using the linear relationship between the local intensity and complex-decorrelation which was termed as inverse SNR-decorrelation (ID) threshold. Furthermore, the ID threshold was automatically determined by defining a binary image similarity (BISIM) index as the feedback and searching the ID threshold with the minimal BISIM value. The proposed ID-BISIM threshold was applied to the acquired OCTA skin images for further vessel quantification.

Results: The proposed ID-BISIM threshold enabled a 3D adaptive binarization and presented superior sensitivity and specificity in vessel segmentation over conventional 2-step thresholding method in the decorrelation-contrast OCTA and a 37-65% improvement of the Youden's index in human skin experiments. The 3D binarization enabled a depth-resolved vessel skeleton and enhanced the differentiation of the overlapping vessels in the depth direction. Using ID-BISIM, the quantitative OCTA image presented a significant increase of vessel diameter index (P=0.0015) and vessel area density (VAD) (P=0.0485) as well as a significant decrease of vessel complexity index (VCI) (P=0.0094) in PWS lesion skin compared with normal skin.

Conclusions: The proposed ID-BISIM method enables an automatic 3D adaptive vessel segmentation with enhanced performance in quantitative OCTA. The vascular quantitative metrics would be a useful tool for improving the diagnosis and the treatment of PWS.

Keywords: Port wine stain (PWS); automatic SNR-adaptive classifier; optical coherence tomography angiography (OCTA); quantification metrics.