Dataset Information

Cardio-pulmonary substructure segmentation of radiotherapy computed tomography images using convolutional neural networks for clinical outcomes analysis.

ABSTRACT:

Background and purpose

Radiation dose to the cardio-pulmonary system is critical for radiotherapy-induced mortality in non-small cell lung cancer. Our goal was to automatically segment substructures of the cardio-pulmonary system for use in outcomes analyses for thoracic cancers. We built and validated a multi-label Deep Learning Segmentation (DLS) model for accurate auto-segmentation of twelve cardio-pulmonary substructures.

Materials and methods

The DLS model utilized a convolutional neural network for segmenting substructures from 217 thoracic radiotherapy Computed Tomography (CT) scans. The model was built in the presence of variable image characteristics such as the absence/presence of contrast. We quantitatively evaluated the final model against expert contours for a hold-out dataset of 24 CT scans using Dice Similarity Coefficient (DSC), 95th Percentile of Hausdorff Distance and Dose-volume Histograms (DVH). DLS contours of an additional 25 scans were qualitatively evaluated by a radiation oncologist to determine their clinical acceptability.

Results

The DLS model reduced segmentation time per patient from about one hour to 10 s. Quantitatively, the highest accuracy was observed for the Heart (median DSC = (0.96 (0.95-0.97)). The median DSC for the remaining structures was between 0.81 and 0.93. No statistically significant difference was found between DVH metrics of the auto-generated and manual contours (p-value ⩾ 0.69). The expert judged that, on average, 85% of contours were qualitatively equivalent to state-of-the-art manual contouring.

Conclusion

The cardio-pulmonary DLS model performed well both quantitatively and qualitatively for all structures. This model has been incorporated into an open-source tool for the community to use for treatment planning and clinical outcomes analysis.

SUBMITTER: Haq R

PROVIDER: S-EPMC7807536 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:PurposeThe deformable nature of the liver can make focal treatment challenging and is not adequately addressed with simple rigid registration techniques. More advanced registration techniques can take deformations into account (eg, biomechanical modeling) but require segmentations of the whole liver for each scan, which is a time-intensive process. We hypothesize that fully convolutional networks can be used to rapidly and accurately autosegment the liver, removing the temporal bottleneck for biomechanical modeling.Methods and materialsManual liver segmentations on computed tomography scans from 183 patients treated at our institution and 30 scans from the Medical Image Computing & Computer Assisted Intervention challenges were collected for this study. Three architectures were investigated for rapid automated segmentation of the liver (VGG-16, DeepLabv3 +, and a 3-dimensional UNet). Fifty-six cases were set aside as a final test set for quantitative model evaluation. Accuracy of the autosegmentations was assessed using Dice similarity coefficient and mean surface distance. Qualitative evaluation was also performed by 3 radiation oncologists on 50 independent cases with previously clinically treated liver contours.ResultsThe mean (minimum-maximum) mean surface distance for the test groups with the final model, DeepLabv3 +, were as follows: μContrast(N = 17): 0.99 mm (0.47-2.2), μNon_Contrast(N = 19)l: 1.12 mm (0.41-2.87), and μMiccai(N = 30)t: 1.48 mm (0.82-3.96). The qualitative evaluation showed that 30 of 50 autosegmentations (60%) were preferred to manual contours (majority voting) in a blinded comparison, and 48 of 50 autosegmentations (96%) were deemed clinically acceptable by at least 1 reviewing physician.ConclusionsThe autosegmentations were preferred compared with manually defined contours in the majority of cases. The ability to rapidly segment the liver with high accuracy achieved in this investigation has the potential to enable the efficient integration of biomechanical model-based registration into a clinical workflow.

Project description:BackgroundPulmonary segments are valuable because they can provide more precise localization and intricate details of lung cancer than lung lobes. With advances in precision therapy, there is an increasing demand for the identification and visualization of pulmonary segments in computed tomography (CT) images to aid in the precise treatment of lung cancer. This study aimed to integrate multiple deep-learning models to accurately segment pulmonary segments in CT images using a bronchial tree (BT)-based approach.MethodsThe proposed segmentation method for pulmonary segments using the BT-based approach comprised the following five essential steps: (I) segmentation of the lung using a U-Net (R231) (public access) model; (II) segmentation of the lobes using a V-Net (self-developed) model; (III) segmentation of the airway using a combination of a differential geometric approach method and a BronchiNet (public access) model; (IV) labeling of the BT branches based on anatomical position; and (V) segmentation of the pulmonary segments based on the distance of each voxel to the labeled BT branches. This five-step process was applied to 14 high-resolution breath-hold CT images and compared against manual segmentations for evaluation.ResultsFor the lung segmentation, the lung mask had a mean dice similarity coefficient (DSC) of 0.98±0.03. For the lobe segmentation, the V-Net model had a mean DSC of 0.94±0.06. For the airway segmentation, the average total length of the segmented airway trees per image scan was 1,902.8±502.1 mm, and the average number of the maximum airway tree generations was 8.5±1.3. For the segmentation of the pulmonary segments, the proposed method had a DSC of 0.73±0.11 and a mean surface distance of 6.1±2.9 mm.ConclusionsThis study demonstrated the feasibility of combining multiple deep-learning models for the auxiliary segmentation of pulmonary segments on CT images using a BT-based approach. The results highlighted the potential of the BT-based method for the semi-automatic segmentation of the pulmonary segment.

Project description:PurposeRadiation dose to the heart and cardiac substructures has been linked to cardiotoxicities. Because cardiac substructures are poorly visualized on treatment-planning computed tomography (CT) scans, we used the superior soft-tissue contrast of magnetic resonance (MR) imaging to optimize a hybrid MR/CT atlas for substructure dose assessment using CT.Methods and materialsThirty-one patients with left-sided breast cancer underwent a T2-weighted MR imaging scan and noncontrast simulation CT scans. A radiation oncologist delineated 13 substructures (chambers, great vessels, coronary arteries, etc) using MR/CT information via cardiac-confined rigid registration. Ground-truth contours for 20 patients were inputted into an intensity-based deformable registration atlas and applied to 11 validation patients. Automatic segmentations involved using majority vote and Simultaneous Truth and Performance Level Estimation (STAPLE) strategies with 1 to 15 atlas matches. Performance was evaluated via Dice similarity coefficient (DSC), mean distance to agreement, and centroid displacement. Three physicians evaluated segmentation performance via consensus scoring by using a 5-point scale. Dosimetric assessment included measurements of mean heart dose, left ventricular volume receiving 5 Gy, and left anterior descending artery mean and maximum doses.ResultsAtlas approaches performed similarly well, with 7 of 13 substructures (heart, chambers, ascending aorta, and pulmonary artery) having DSC >0.75 when averaged over 11 validation patients. Coronary artery segmentations were not successful with the atlas-based approach (mean DSC <0.3). The STAPLE method with 10 matches yielded the highest DSC and the lowest mean distance to agreement for all high-performing substructures (omitting coronary arteries). For the STAPLE method with 10 matches, >50% of all validation contours had centroid displacements <3.0 mm, with the largest shifts in the coronary arteries. Atlas-generated contours had no statistical difference from ground truth for left anterior descending artery maximum dose, mean heart dose, and left ventricular volume receiving 5 Gy (P > .05). Qualitative contour grading showed that 8 substructures required minor modifications.ConclusionsThe hybrid MR/CT atlas provided reliable segmentations of chambers, heart, and great vessels for patients undergoing noncontrast CT, suggesting potential widespread applicability for routine treatment planning.