Automatic segmentation of multiple cardiovascular structures from cardiac computed tomography angiography images using deep learning

• Type of Material: Article
• Authors: Lohendran Baskaran, Weill Cornell Medical College; Subhi J. Al'Aref, Weill Cornell Medical College; Gabriel Maliakal, Cleerly Inc.; Benjamin C. Lee, Weill Cornell Medical College; Zhuoran Xu, Weill Cornell Medical College; Jeong W. Choi, Weill Cornell Medical College; Sang-Eun Lee, Yonsei University; Ji Min Sung, Yonsei University; Fay Y. Lin, Weill Cornell Medical College; Simon Dunham, Weill Cornell Medical College; Bobak Mosadegh, Weill Cornell Medical College; Yong-Jin Kim, Seoul National University; Ilan Gottlieb, Casa de Saude Sao Jose; Byoung Kwon Lee, Yonsei University; Eun Ju Chun, Seoul National University; Filippo Cademartiri, SDN IRCCS; Erica Maffei, ASUR Marche; Hugo Marques, Hospital da Luz, Lisboa, Portugal; Sanghoon Shin, Ewha Womans University; Jung Hyun Choi, Pusan University Hospital; Kavitha Chinnaiyan, William Beaumont Hospital; Martin Hadamitzky, German Heart Center Munich; Edoardo Conte, IRCCS; Daniele Andreini, IRCCS; Gianluca Pontone, IRCCS; Matthew J. Budoff, Los Angeles Biomedical Research Institute; Jonathon A. Leipsic, University of British Columbia; Gilbert L. Raff, William Beaumont Hospital; Renu Virmani, William Beaumont Hospital; Habib Samady, Emory University; Peter H. Stone, Harvard Medical School; Daniel S. Berman, Cedars Sinai Medical Center; Jagat Narula, Icahn School of Medicine at Mount Sinai; Jeroen J. Bax, Leiden University; Hyuk-Jae Chang, Yonsei University; James K. Min, Weill Cornell Medical College; Leslee Shaw, Emory University
• Language: English
• Date: 2020-05-06
• Publisher: Public Library Science
• Publication Version: Final Published Version
• Copyright Statement: © 2020 Baskaran et al.
• License: Creative Commons Attribution 4.0 International
• Final Published Version (URL): http://dx.doi.org/10.1371/journal.pone.0232573
• Title of Journal or Parent Work: PLoS ONE
• Volume: 15
• Issue: 5
• Start Page: e0232573
• End Page: e0232573
• Grant/Funding Information: The research reported in this manuscript was supported by the Dalio Institute of Cardiovascular Imaging (New York, NY, USA).; Dr. Jonathon Leipsic is a consultant and has reported stock options with Circle CVI and HeartFlow.; James K. Min received funding from the Dalio Foundation, National Institutes of Health, and GE Healthcare.; Dr. Min serves on the scientific advisory board of Arineta and GE Healthcare, and became founder and an employee of Cleerly, Inc after this research was conducted.
• Keywords: Automatic segmentation; Cardiovascular structures; Science & Technology; Multidisciplinary Sciences; Heart; Science & Technology - Other Topics; Computed tomography
• Research Categories: Health Sciences, Pathology; Health Sciences, Radiology; Engineering, Biomedical

Abstract

Objectives: To develop, demonstrate and evaluate an automated deep learning method for multiple cardiovascular structure segmentation. Background: Segmentation of cardiovascular images is resource-intensive. We design an automated deep learning method for the segmentation of multiple structures from Coronary Computed Tomography Angiography (CCTA) images. Methods: Images from a multicenter registry of patients that underwent clinically-indicated CCTA were used. The proximal ascending and descending aorta (PAA, DA), superior and inferior vena cavae (SVC, IVC), pulmonary artery (PA), coronary sinus (CS), right ventricular wall (RVW) and left atrial wall (LAW) were annotated as ground truth. The U-net-derived deep learning model was trained, validated and tested in a 70:20:10 split. Results: The dataset comprised 206 patients, with 5.130 billion pixels. Mean age was 59.9 ± 9.4 yrs., and was 42.7% female. An overall median Dice score of 0.820 (0.782, 0.843) was achieved. Median Dice scores for PAA, DA, SVC, IVC, PA, CS, RVW and LAW were 0.969 (0.979, 0.988), 0.953 (0.955, 0.983), 0.937 (0.934, 0.965), 0.903 (0.897, 0.948), 0.775 (0.724, 0.925), 0.720 (0.642, 0.809), 0.685 (0.631, 0.761) and 0.625 (0.596, 0.749) respectively. Apart from the CS, there were no significant differences in performance between sexes or age groups. Conclusions: An automated deep learning model demonstrated segmentation of multiple cardiovascular structures from CCTA images with reasonable overall accuracy when evaluated on a pixel level.