Automatic delineation of cardiac substructures using a region‐based fully convolutional network

Abstract

Purpose Radiation dose to specific cardiac substructures, such as the atria and ventricles, has been linked to post‐treatment toxicity and has shown to be more predictive of these toxicities than dose to the whole heart. A deep learning‐based algorithm for automatic generation of these contours is proposed to aid in either retrospective or prospective dosimetric studies to better understand the relationship between radiation dose and toxicities. Methods The proposed method uses a mask‐scoring regional convolutional neural network (RCNN) which consists of five major subnetworks: backbone, regional proposal network (RPN), RCNN head, mask head, and mask‐scoring head. Multiscale feature maps are learned from computed tomography (CT) via the backbone network. The RPN utilizes these feature maps to detect the location and region‐of‐interest (ROI) of all substructures, and the final three subnetworks work in series to extract structural information from these ROIs. The network is trained using 55 patient CT datasets, with 22 patients having contrast scans. Threefold cross validation (CV) is used for evaluation on 45 datasets, and a separate cohort of 10 patients are used for holdout evaluation. The proposed method is compared to a 3D UNet. Results The proposed method produces contours that are qualitatively similar to the ground truth contours. Quantitatively, the proposed method achieved average Dice score coefficients (DSCs) for the whole heart, chambers, great vessels, coronary arteries, the valves of the heart of 0.96, 0.94, 0.93, 0.66, and 0.77 respectively, outperforming the 3D UNet, which achieved DSCs of 0.92, 0.87, 0.88, 0.48, and 0.59 for the corresponding substructure groups. Mean surface distances (MSDs) between substructures segmented by the proposed method and the ground truth were <2 mm except for the left anterior descending coronary artery and the mitral and tricuspid valves, and <5 mm for all substructures. When dividing results into noncontrast and contrast datasets, the model performed statistically significantly better in terms of DSC, MSD, centroid mean distance (CMD), and volume difference for the chambers and whole heart with contrast. Notably, the presence of contrast did not statistically significantly affect coronary artery segmentation DSC or MSD. After network training, all substructures and the whole heart can be segmented on new datasets in less than 5 s. Conclusions A deep learning network was trained for automatic delineation of cardiac substructures based on CT alone. The proposed method can be used as a tool to investigate the relationship between cardiac substructure dose and treatment toxicities.

The impact of organ-at-risk contour variations on automatically generated treatment plans for NSCLC

“…Recently, there has been a large interest in substructures of the heart and identification of the DVH-parameters of these substructures to patient outcome [31][32][33][34]. However, it is still not entirely clear which substructures have largest impact.…”

Automatic Cardiac Structure Contouring for Small Datasets with Cascaded Deep Learning Models

“…In the case of atlas based methods, N indicates the number of volumes used to create the atlas/the number tested on. Bold values indicate the highest score of all the NCCT projects Choi et al [ 15 ] (N = 35/14) (NCCT) (DL based) Harms et al [ 16 ] (N = 30/15) (CCT & NCCT) (DL based) Jung et al [ 17 ] (N = 29/1) (NCCT) (Atlas based) Luo et al [ 18 ] (N = 12/49) (NCCT) (Atlas based) Proposed method (N = 41/6) (NCCT) (DL based) DC 95% HD (mm) DC 95% HD (mm) DC 95% HD (mm) DC 95% HD (mm) DC 95% HD (mm) WH 0.95 ± 0.01 2.39 ± 0.47 0.96 ± 0.03 6.00 ± 5.73 0.97 ± 0.01 NR 0.95 ± 0.04 NR 0.95 ± 0.01 2.22 ± 0.97 RV 0.86 ± 0.04 6.15 ± 2.39 0.92 ± 0.03 4.43 ± 1.95 0.69 ± 0.08 NR 0.87 ± 0.10 NR 0.88 ± 0.02 2.88 ± 0.67 LV 0.87 ± 0.03 4.52 ± 1.65 0.96 ± 0.01 3.80 ± 1.46 0.80 ± 0.06 NR 0.91 ± 0.06 NR 0.92 ± 0.01 3.27 ±...…”

Deep learning model for automatic contouring of cardiovascular substructures on radiotherapy planning CT images: Dosimetric validation and reader study based clinical acceptability testing

“…Other models such as those trained with images from breast cancer patients [16] are unlikely to generalize as well for lung cancer patient images which regularly contain tumors close to the heart. Despite the current state of the art for automatic cardiac contouring in lung cancer patients reporting high accuracies for the whole heart, the cardiac chambers and the great vessels [17,18], it still lacks in clinical validation. Specifically, model accuracy has been evaluated in datasets of only a few dozens of patients, often from a single institute and dosimetric parameter estimation and clinical acceptability has either not been evaluated [17], or only evaluated at the same scale [18].…”

“…Despite the current state of the art for automatic cardiac contouring in lung cancer patients reporting high accuracies for the whole heart, the cardiac chambers and the great vessels [17,18], it still lacks in clinical validation. Specifically, model accuracy has been evaluated in datasets of only a few dozens of patients, often from a single institute and dosimetric parameter estimation and clinical acceptability has either not been evaluated [17], or only evaluated at the same scale [18]. Impact of adjacency of the tumor to the contoured structures has also not been reported.…”