Xiaowei Ding1, Demetri Terzopoulos2, Mariana Diaz-Zamudio3, Daniel S Berman4, Piotr J Slomka4, Damini Dey5. 1. Biomedical Imaging Research Institute, Department of Biomedical Sciences, Cedars Sinai Medical Center, Los Angeles, California 90048 and Computer Science Department, Henry Samueli School of Engineering and Applied Science at UCLA, Los Angeles, California 90095. 2. Computer Science Department, Henry Samueli School of Engineering and Applied Science at UCLA, Los Angeles, California 90095. 3. Nuclear Medicine Department, Cedars Sinai Medical Center, Los Angeles, California 90048. 4. Departments of Imaging and Medicine, Cedars Sinai Medical Center, Los Angeles, California 90048 and Department of Medicine, David-Geffen School of Medicine at UCLA, Los Angeles, California 90095. 5. Biomedical Imaging Research Institute, Department of Biomedical Sciences, Cedars Sinai Medical Center, Los Angeles, California 90048 and Department of Medicine, David-Geffen School of Medicine at UCLA, Los Angeles, California 90095.
Abstract
PURPOSE: The authors aimed to develop and validate an automated algorithm for epicardial fat volume (EFV) quantification from noncontrast CT. METHODS: The authors developed a hybrid algorithm based on initial segmentation with a multiple-patient CT atlas, followed by automated pericardium delineation using geodesic active contours. A coregistered segmented CT atlas was created from manually segmented CT data and stored offline. The heart and pericardium in test CT data are first initialized by image registration to the CT atlas. The pericardium is then detected by a knowledge-based algorithm, which extracts only the membrane representing the pericardium. From its initial atlas position, the pericardium is modeled by geodesic active contours, which iteratively deform and lock onto the detected pericardium. EFV is automatically computed using standard fat attenuation range. RESULTS: The authors applied their algorithm on 50 patients undergoing routine coronary calcium assessment by CT. Measurement time was 60 s per-patient. EFV quantified by the algorithm (83.60 ± 32.89 cm(3)) and expert readers (81.85 ± 34.28 cm(3)) showed excellent correlation (r = 0.97, p < 0.0001), with no significant differences by comparison of individual data points (p = 0.15). Voxel overlap by Dice coefficient between the algorithm and expert readers was 0.92 (range 0.88-0.95). The mean surface distance and Hausdorff distance in millimeter between manually drawn contours and the automatically obtained contours were 0.6 ± 0.9 mm and 3.9 ± 1.7 mm, respectively. Mean difference between the algorithm and experts was 9.7% ± 7.4%, similar to interobserver variability between 2 readers (8.0% ± 5.3%, p = 0.3). CONCLUSIONS: The authors' novel automated method based on atlas-initialized active contours accurately and rapidly quantifies EFV from noncontrast CT.
PURPOSE: The authors aimed to develop and validate an automated algorithm for epicardial fat volume (EFV) quantification from noncontrast CT. METHODS: The authors developed a hybrid algorithm based on initial segmentation with a multiple-patient CT atlas, followed by automated pericardium delineation using geodesic active contours. A coregistered segmented CT atlas was created from manually segmented CT data and stored offline. The heart and pericardium in test CT data are first initialized by image registration to the CT atlas. The pericardium is then detected by a knowledge-based algorithm, which extracts only the membrane representing the pericardium. From its initial atlas position, the pericardium is modeled by geodesic active contours, which iteratively deform and lock onto the detected pericardium. EFV is automatically computed using standard fat attenuation range. RESULTS: The authors applied their algorithm on 50 patients undergoing routine coronary calcium assessment by CT. Measurement time was 60 s per-patient. EFV quantified by the algorithm (83.60 ± 32.89 cm(3)) and expert readers (81.85 ± 34.28 cm(3)) showed excellent correlation (r = 0.97, p < 0.0001), with no significant differences by comparison of individual data points (p = 0.15). Voxel overlap by Dice coefficient between the algorithm and expert readers was 0.92 (range 0.88-0.95). The mean surface distance and Hausdorff distance in millimeter between manually drawn contours and the automatically obtained contours were 0.6 ± 0.9 mm and 3.9 ± 1.7 mm, respectively. Mean difference between the algorithm and experts was 9.7% ± 7.4%, similar to interobserver variability between 2 readers (8.0% ± 5.3%, p = 0.3). CONCLUSIONS: The authors' novel automated method based on atlas-initialized active contours accurately and rapidly quantifies EFV from noncontrast CT.
Authors: Leon Lenchik; Laura Heacock; Ashley A Weaver; Robert D Boutin; Tessa S Cook; Jason Itri; Christopher G Filippi; Rao P Gullapalli; James Lee; Marianna Zagurovskaya; Tara Retson; Kendra Godwin; Joey Nicholson; Ponnada A Narayana Journal: Acad Radiol Date: 2019-08-10 Impact factor: 3.173
Authors: Alexander Norlén; Jennifer Alvén; David Molnar; Olof Enqvist; Rauni Rossi Norrlund; John Brandberg; Göran Bergström; Fredrik Kahl Journal: J Med Imaging (Bellingham) Date: 2016-09-15
Authors: Frederic Commandeur; Markus Goeller; Julian Betancur; Sebastien Cadet; Mhairi Doris; Xi Chen; Daniel S Berman; Piotr J Slomka; Balaji K Tamarappoo; Damini Dey Journal: IEEE Trans Med Imaging Date: 2018-02-09 Impact factor: 10.048
Authors: Frederic Commandeur; Markus Goeller; Aryabod Razipour; Sebastien Cadet; Michaela M Hell; Jacek Kwiecinski; Xi Chen; Hyuk-Jae Chang; Mohamed Marwan; Stephan Achenbach; Daniel S Berman; Piotr J Slomka; Balaji K Tamarappoo; Damini Dey Journal: Radiol Artif Intell Date: 2019-11-27