J Yang

Cardiac atlas development and validation for automatic segmentation of cardiac substructures

PURPOSE To develop and validate a set of atlases for auto-contouring cardiac substructures. METHODS Eight radiation oncologists manually and independently delineated 15 cardiac substructures from noncontrast CT images of 6 patients by referring to their respective fused contrast CT images. Individual contours were fused together for each structure, edited by 2 physicians, and became atlases to delineate other 6 patients. The auto-delineated contours of the 6 additional patients became templates for manual contouring. These 12 patients with well-defined contours composed the final atlases for multi-atlas segmentation. RESULTS The average time for manually contouring the 15 cardiac substructures was about 40min. Inter-observer variability was small for the heart, the chambers, and the aorta compared with that for other structures that were not clearly distinguishable in CT images. The mean dice similarity coefficient and mean surface distance of auto-segmented contours were within one standard deviation of expert contouring variability. Good agreement between auto-segmented and manual contours was observed for the heart, the chambers, and the great vessels. Independent validation on other 19 patients showed reasonable agreement for the heart chambers. CONCLUSIONS A set of cardiac atlases was created for auto-contouring from noncontrast CT images. The accuracy of auto-contouring for the heart, chambers, and great vessels was validated for potential clinical use.

SU-E-J-261: The Importance of Appropriate Image Preprocessing to Augment the Information of Radiomics Image Features

Purpose: To investigate how different image preprocessing techniques, their parameters, and the different boundary handling techniques can augment the information of features and improve feature’s differentiating capability. Methods: Twenty-seven NSCLC patients with a solid tumor volume and no visually obvious necrotic regions in the simulation CT images were identified. Fourteen of these patients had a necrotic region visible in their pre-treatment PET images (necrosis group), and thirteen had no visible necrotic region in the pre-treatment PET images (non-necrosis group). We investigated how image preprocessing can impact the ability of radiomics image features extracted from the CT to differentiate between two groups. It is expected the histogram in the necrosis group is more negatively skewed, and the uniformity from the necrosis group is less. Therefore, we analyzed two first order features, skewness and uniformity, on the image inside the GTV in the intensity range [−20HU, 180HU] under the combination of several image preprocessing techniques: (1) applying the isotropic Gaussian or anisotropic diffusion smoothing filter with a range of parameter(Gaussian smoothing: size=11, sigma=0:0.1:2.3; anisotropic smoothing: iteration=4, kappa=0:10:110); (2) applying the boundaryadapted Laplacian filter; and (3) applying the adaptive upper threshold for the intensity range. A 2-tailed T-test was used to evaluate the differentiating capability of CT features on pre-treatment PT necrosis. Result: Without any preprocessing, no differences in either skewness or uniformity were observed between two groups. After applying appropriate Gaussian filters (sigma>=1.3) or anisotropic filters(kappa >=60) with the adaptive upper threshold, skewness was significantly more negative in the necrosis group(p<0.05). By applying the boundary-adapted Laplacian filtering after the appropriate Gaussian filters (0.5 <=sigma<=1.1) or anisotropic filters(20<=kappa <=50), the uniformity was significantly lower in the necrosis group (p<0.05). Conclusion: Appropriate selection of image preprocessing techniques allows radiomics features to extract more useful information and thereby improve prediction models based on these features.

SU‐D‐144‐03: Respiratory Motion Management for High‐Precision Small Animal Irradiation

PURPOSE The ability to produce precisely targeted beams as small as 1 mm necessitates the understanding and management of intra-fraction motion. This study evaluated lung motion in free-breathing mice and compared free-breathing imaging (3D and 4D reconstruction) to breath-hold imaging for use in treatment planning. METHODS Five mice were imaged weekly for six weeks using the X-RAD 225Cx system. Each week, CBCT projections were acquired during free-breathing imaging under anesthesia and reconstructed into 3D and 4D images (6 phases). Superior-inferior, anterior-posterior, and right-left motion were evaluated using deformable registration of the 4D images, and confirmed by manual measurements on fluoroscopic images. Next, the mice were intubated and their breath was held at full-inhale for 20 seconds during image acquisition. Breath-hold scans from the same session were compared to assess reproducibility. RESULTS The average voxel motion in the lungs of free-breathing mice was 1.3 mm (stdev = 0.2mm) while the average maximum motion was 3.4 mm (stdev = 0.3mm). To ensure tumor coverage in free-breathing mice we can apply the ITV concept to 3D-CBCTs, using added margins defined by the 4D-CBCT image sets. However, in an area of maximum motion, the expanded target volume could cover 30% the length of the lung. Adding this margin could Result in substantial normal tissue toxicity. Breath-hold imaging was reproducible to within 0.6 mm except in areas close to the heart due to cardiac motion. A visual comparison of image quality found that the breath-hold images were substantially sharper than both the 3D and 4D free-breathing images due to blurring and reconstruction artifacts. CONCLUSION Given the large respiratory motion relative to lung size in mice, breath-hold imaging offers clear advantages in motion management and image quality. Breath-hold treatments using IGRT are feasible and are recommended in cases of large tumor motion.

The Pulmonary Fibrosis Associated MUC5B Promoter Polymorphism Is Prognostic of the Overall Survival in Patients with Non–Small Cell Lung Cancer (NSCLC) Receiving Definitive Radiotherapy

BACKGROUND: MUC5B is glycoprotein secreted by bronchial glands. A promoter variant in MUC5B, rs35705950, was previously found to be strongly associated with the incidence of idiopathic pulmonary fibrosis (IPF) and also the overall survival (OS) of such patients. Patients with IPF and patients with radiation pneumonitis (RP) have the similar pathologic process and clinical symptoms. However, the role of rs35705950 in patients receiving thoracic radiotherapy remains unclear. PATIENTS AND METHODS: In total, 664 patients with NSCLC receiving definitive radiotherapy (total dose ≥60 Gy) were included in our study. RP was scored via the Common Terminology Criteria for Adverse Events v3.0. OS was the second end point. MUC5B rs35705950 was genotyped, and Kaplan-Meier and Cox regression analyses were used to evaluate associations between MUC5B rs35705950 and the risk of RP or OS. RESULTS: The median patient age was 66 years (range 35-88); most (488 [73.2%]) had stage III of the disease. Until the last follow-up, 250 patients developed grade ≥ 2 RP, 82 patients developed grade ≥ 3 RP, and 440 patients died. The median mean lung dose was 17.9 Gy (range 0.15-32.74). No statistically significant associations were observed between genotypes of MUC5B rs35705950 and the incidence of RP ≥ grade 2 either in univariate analysis (hazard ratio [HR] 1.009, 95% confidence interval [CI] 0.728-1.399, P = .958) or in multivariate analysis (HR 0.921, 95% CI 0.645-1.315, P = .65). Similar results were also observed for RP ≥ grade 3, while TT/GT genotypes in MUC5B were significantly associated with poor OS in both univariate analysis (HR 1.287, 95% CI 1.009-1.640, P = .042) and multivariate analysis (HR 1.561, 95% CI 1.193-2.042, P = .001). CONCLUSION: MUC5B promoter polymorphism could be prognostic of the OS among NSCLC patients receiving definitive radiotherapy, although no significant associations were found with the risk of RP.

SU-F-R-09: Homogenization of CT Images for Radiomics Studies: It’s Like Butter(worth)

PURPOSE Previous studies have shown that differences in the CT image acquisition parameters produce variability in extracted quantitative image features. The purpose of our study was to determine if the variability due to inconsistencies in the reconstruction field of view (FOV), or pixel size, can be reduced by image filtering. METHODS We reconstructed the CT scans of 8 NSCLC patients 5 times, varying the FOV from 30 to 50 cm. We then calculated 150 radiomics features for each of the 40 CT image sets using 6 preprocessing methods. The preprocessing methods included re-sampling the images to a consistent 1 mm per pixel and Butterworth filters with cutoff frequencies from 75 to 200. We next calculated the overall concordance correlation coefficient (OCCC) for each feature to assess the intra-patient variability due to multiple FOVs relative to the inter-patient variability. To further explore how image preprocessing can reduce the effects of the reconstruction FOV, we performed hierarchical clustering. RESULTS 99% of all features had an OCCC of greater than 0.95 for images resampled and processed with a Butterworth filter with cutoff frequency 125, compared to only 21% of the features calculated without image preprocessing. Filtered images had consistently larger OCCC values except for shape related features which had all OCCC values > 0.95 regardless of the image preprocessing. Without filtering, hierarchical clustering was able to correctly group the 5 FOV scans for only 2 of the 8 patients. For images resample and processed with a Butterworth filter with cutoff frequency 125, the hierarchical clustering correctly grouped the 5 FOV scans for all 8 patients. CONCLUSION Radiomics features, especially those that relate the image intensity and spatial information, show a strong dependence on the reconstruction FOV. However, our results demonstrate that resampling and filtering the images can greatly reduce this dependence.

SU‐E‐J‐256: Dual Energy Planar Image Based Localization in the Absence of On‐Board CT Images

PURPOSE To develop a tool enabling soft tissue based image guidance using dual energy radiographs for cases when on-board CT is not available. METHOD Dual energy planar radiographs can be applied to image guidance for targeting lung lesions because the bone based alignment only may not be sufficient as the lesions move. We acquired images of an anthropomorphic thorax phantom at 120 and 60 KVp respectively. Using a weighted logarithmic subtraction of these dual energy images, a soft tissue enhanced and a bone enhanced image were generated and they could be used for the image guidance purpose. Similar processing was also applied to a dual energy image set acquired for a patient undergoing a proton therapy. RESULTS The soft tissue enhanced images suppressed bones (ribs and scapula) overlying on lung, thus enabling a better visualization of soft tissue and lesion, while the bone enhanced image suppressed the soft tissue. These enhanced effects were visually apparent without further processing for display enhancements, such as using histogram or edge enhancement technique. CONCLUSIONS The phantom image processing was encouraging. The initial test on the patient image set showed that other post processing might still be able to add value in visualizing soft tissues in addition to the dual energy soft tissue enhancement. More evaluations are needed to determine the potential benefit of this technique in the clinic.

SU‐E‐T‐359: Patients Could (and Should) Be Treated in An Upright Position

PURPOSE Treating patients in an upright (seated) position has several potential advantages including increased lung volume/reduced respiratory motion (lung patients) giving an improved lung DVH, moving other healthy tissue away from the treatment beams (e.g. breast for Hodgkins lymphoma patients), and improved comfort (H&N patients with excess mucus secretion, patients with compromised lung function, etc.). We report on efforts to develop the tools/techniques to add upright treatments to the modern radiotherapy armory. METHODS The following steps have been taken towards treating patients in an upright position: (1) A novel MRI scanner that allows scans to be taken at any angle between vertical and horizontal positions was used to take volume scans and sagittal movies of 5 volunteers in vertical and supine positions. Lung volumes and motion were compared. (2) An innovative treatment chair and immobilization approach has been prototyped. The chair attaches to the treatment couch, allowing full use of the LINACs on-board imaging and remote motion functions. (3) kV cone-beam CT images have been acquired of upright phantoms using the Truebeam with the gantry at 0degrees and rotating the couch instead of the gantry. (4) Treatment planning studies have been completed for H&N and thoracic treatments. RESULTS (1) Lung volume is increased by 6-53% and motion reduced by 1-4mm, when patients are positioned in an upright position. (2) The chair has good comfort and immobilization. End-of-couch weight-constraint is sufficient with the TrueBeam couch, but limits use for older couches. (3) Upright CBCTs are possible, although work is needed to improve the image quality. (4) We have quantified gantry/patient collision risks, and understand the gantry/couch angle combinations that will be used for treatment. CONCLUSION Good progress has been made in developing the techniques/tools to treat patients in an upright position. Partially funded by Varian Medical Systems.

Polymorphisms in BMP2/BMP4, with estimates of mean lung dose, predict radiation pneumonitis among patients receiving definitive radiotherapy for non-small cell lung cancer

Single nucleotide polymorphisms (SNPs) in TGFβ1 can predict the risk of radiation pneumonitis (RP) in patients with non-small cell lung cancer (NSCLC) after definitive radiotherapy. Here we investigated whether SNPs in TGFβ superfamily members BMP2 and BMP4 are associated with RP in such patients. In total, we retrospectively analyzed 663 patients given ≥ 60 Gy for NSCLC. We randomly assigned 323 patients to the training cohort and 340 patients to the validation cohort. Potentially functional and tagging SNPs of BMP2 (rs170986, rs1979855, rs1980499, rs235768, rs3178250) and BMP4 (rs17563, rs4898820, rs762642) were genotyped. The median of mean lung dose (MLD) was 17.9 Gy (range, 0.15–32.74 Gy). Higher MLD was strongly associated with increased risk of grade ≥ 2 RP (hazard ratio [HR]=2.191, 95% confidence interval [CI] = 1.680–2.856, P < 0.001) and grade ≥ 3 RP (HR = 4.253, 95% CI = 2.493–7.257, P < 0.001). In multivariate analyses, BMP2 rs235768 AT/TT was associated with higher risk of grade ≥ 2 RP (HR = 1.866, 95% CI = 1.221–2.820, P = 0.004 vs. AA) both in training cohort and validation cohort. Similar results were observed for BMP2 rs1980499. BMP2 rs3178250 CT/TT was associated with lower risk of grade ≥ 3 RP (HR = 0.406, 95% CI = 0.175–0.942, P = 0.036 vs. CC) in the pooled analysis. Adding the rs235768 and rs1980499 SNPs to a model comprising age, performance status, and MLD raised the Harrells C for predicting grade ≥ 2 RP from 0.6117 to 0.6235 (P = 0.0105). SNPs in BMP2 can predict grade ≥ 2 or 3 RP after radiotherapy for NSCLC and improve the predictive power of MLD model. Validation is underway through an ongoing prospective trial.

TU-H-CAMPUS-JeP1-02: Fully Automatic Verification of Automatically Contoured Normal Tissues in the Head and Neck

PURPOSE To investigate and validate the use of an independent deformable-based contouring algorithm for automatic verification of auto-contoured structures in the head and neck towards fully automated treatment planning. METHODS Two independent automatic contouring algorithms [(1) Eclipses Smart Segmentation followed by pixel-wise majority voting, (2) an in-house multi-atlas based method] were used to create contours of 6 normal structures of 10 head-and-neck patients. After rating by a radiation oncologist, the higher performing algorithm was selected as the primary contouring method, the other used for automatic verification of the primary. To determine the ability of the verification algorithm to detect incorrect contours, contours from the primary method were shifted from 0.5 to 2cm. Using a logit model the structure-specific minimum detectable shift was identified. The models were then applied to a set of twenty different patients and the sensitivity and specificity of the models verified. RESULTS Per physician rating, the multi-atlas method (4.8/5 point scale, with 3 rated as generally acceptable for planning purposes) was selected as primary and the Eclipse-based method (3.5/5) for verification. Mean distance to agreement and true positive rate were selected as covariates in an optimized logit model. These models, when applied to a group of twenty different patients, indicated that shifts could be detected at 0.5cm (brain), 0.75cm (mandible, cord), 1cm (brainstem, cochlea), or 1.25cm (parotid), with sensitivity and specificity greater than 0.95. If sensitivity and specificity constraints are reduced to 0.9, detectable shifts of mandible and brainstem were reduced by 0.25cm. These shifts represent additional safety margins which might be considered if auto-contours are used for automatic treatment planning without physician review. CONCLUSION Automatically contoured structures can be automatically verified. This fully automated process could be used to flag auto-contours for special review or used with safety margins in a fully automatic treatment planning system.

TU-H-CAMPUS-JeP2-05: Can Automatic Delineation of Cardiac Substructures On Noncontrast CT Be Used for Cardiac Toxicity Analysis?

PURPOSE To evaluate the feasibility of using an automatic segmentation tool to delineate cardiac substructures from computed tomography (CT) images for cardiac toxicity analysis for non-small cell lung cancer (NSCLC) patients after radiotherapy. METHODS A multi-atlas segmentation tool developed in-house was used to delineate eleven cardiac substructures including the whole heart, four heart chambers, and six greater vessels automatically from the averaged 4DCT planning images for 49 NSCLC patients. The automatic segmented contours were edited appropriately by two experienced radiation oncologists. The modified contours were compared with the auto-segmented contours using Dice similarity coefficient (DSC) and mean surface distance (MSD) to evaluate how much modification was needed. In addition, the dose volume histogram (DVH) of the modified contours were compared with that of the auto-segmented contours to evaluate the dosimetric difference between modified and auto-segmented contours. RESULTS Of the eleven structures, the averaged DSC values ranged from 0.73 ± 0.08 to 0.95 ± 0.04 and the averaged MSD values ranged from 1.3 ± 0.6 mm to 2.9 ± 5.1mm for the 49 patients. Overall, the modification is small. The pulmonary vein (PV) and the inferior vena cava required the most modifications. The V30 (volume receiving 30 Gy or above) for the whole heart and the mean dose to the whole heart and four heart chambers did not show statistically significant difference between modified and auto-segmented contours. The maximum dose to the greater vessels did not show statistically significant difference except for the PV. CONCLUSION The automatic segmentation of the cardiac substructures did not require substantial modification. The dosimetric evaluation showed no statistically significant difference between auto-segmented and modified contours except for the PV, which suggests that auto-segmented contours for the cardiac dose response study are feasible in the clinical practice with a minor modification to the PV vessel.