(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: evaluate radiotherapy cardiotoxicity, an accurate cardiac substructure segmentation on radiotherapy planning CT (pCT) is necessity. This work is aimed to develop an image-modality free semi-automatic cardiac substructure segmentation platform on non-contrast pCT using atlas and landmark combined technique.
Methods: front-end of the platform is designed to guide physicians to draw regions of interest to derive 11 landmarks on pCT near apex, aortic/pulmonary valve, mitral/tricuspid valve, entrance of the superior/inferior vena cava, entrance of the right superior/inferior pulmonary vein, and entrance of the left superior/inferior pulmonary vein. These locations are picked as image visual and mathematic transformation stable. The back-end of the platform consists of an atlas and a contour propagation algorithm. A contrast-enhanced cardiac CT (cCT) is used as the atlas image with 7 cardiac substructures, including left/right ventricular cavity (LV, RV), left/right atrial cavity (LA, RA), myocardium of the left ventricle (Myo), ascending aorta (AA), and pulmonary artery (PA) delineated and 11 landmarks drawn. The contour propagation algorithm first conducts pCT’s and cCT’s landmarks rigid alignment via a three-dimensional normal-distributions transform. Then a thin-plate splines transformation (TPS) is calculated on the rigid-aligned landmarks. Finally, the obtained TPS is used to propagate 7 cardiac substructures from cCT to pCT.
Results: to the availability of ground-truth contours, the initial evaluation of platform performance was conducted on 5 cCT cases provided by MICCAI 2017 Multi-Modality Whole Heart Segmentation (MM-WHS). Dice similarity coefficient was used to quantify segmentation accuracy. Evaluation results show the platform achieves DSC 75.42%±9.88%, 89.22%±3.98%, 79.85%±13.66%, 74.46%±3.73%, 84.43%±11.27%, 66.56%±8.46%, 52.90%±7.74% of Myo, LA, LV, RA, RV, AA, and PA, respectively.
Conclusion: developed semi-automatic segmentation platform, which is image modality independent, yields accurate cardiac substructures delineation in evaluation cases. This accurate result indicate it the developed platform is promising for cardiotoxicity evaluation in radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by NIH R01 CA235723 and American Heart Association Grant (No. 19AMTG35120501)
Keywords
Taxonomy
Contact Email
