J Oliver

Study of 201 non-small cell lung cancer patients given stereotactic ablative radiation therapy shows local control dependence on dose calculation algorithm.

PURPOSE Pencil beam (PB) and collapsed cone convolution (CCC) dose calculation algorithms differ significantly when used in the thorax. However, such differences have seldom been previously directly correlated with outcomes of lung stereotactic ablative body radiation (SABR). METHODS AND MATERIALS Data for 201 non-small cell lung cancer patients treated with SABR were analyzed retrospectively. All patients were treated with 50 Gy in 5 fractions of 10 Gy each. The radiation prescription mandated that 95% of the planning target volume (PTV) receive the prescribed dose. One hundred sixteen patients were planned with BrainLab treatment planning software (TPS) with the PB algorithm and treated on a Novalis unit. The other 85 were planned on the Pinnacle TPS with the CCC algorithm and treated on a Varian linac. Treatment planning objectives were numerically identical for both groups. The median follow-up times were 24 and 17 months for the PB and CCC groups, respectively. The primary endpoint was local/marginal control of the irradiated lesion. Grays competing risk method was used to determine the statistical differences in local/marginal control rates between the PB and CCC groups. RESULTS Twenty-five patients planned with PB and 4 patients planned with the CCC algorithms to the same nominal doses experienced local recurrence. There was a statistically significant difference in recurrence rates between the PB and CCC groups (hazard ratio 3.4 [95% confidence interval: 1.18-9.83], Grays test P=.019). The differences (Δ) between the 2 algorithms for target coverage were as follows: ΔD99GITV = 7.4 Gy, ΔD99PTV = 10.4 Gy, ΔV90GITV = 13.7%, ΔV90PTV = 37.6%, ΔD95PTV = 9.8 Gy, and ΔDISO = 3.4 Gy. GITV = gross internal tumor volume. CONCLUSIONS Local control in patients receiving who were planned to the same nominal dose with PB and CCC algorithms were statistically significantly different. Possible alternative explanations are described in the report, although they are not thought likely to explain the difference. We conclude that the difference is due to relative dosimetric underdosing of tumors with the PB algorithm.

Variability of Image Features Computed from Conventional and Respiratory-Gated PET/CT Images of Lung Cancer.

Radiomics is being explored for potential applications in radiation therapy. How various imaging protocols affect quantitative image features is currently a highly active area of research. To assess the variability of image features derived from conventional [three-dimensional (3D)] and respiratory-gated (RG) positron emission tomography (PET)/computed tomography (CT) images of lung cancer patients, image features were computed from 23 lung cancer patients. Both protocols for each patient were acquired during the same imaging session. PET tumor volumes were segmented using an adaptive technique which accounted for background. CT tumor volumes were delineated with a commercial segmentation tool. Using RG PET images, the tumor center of mass motion, length, and rotation were calculated. Fifty-six image features were extracted from all images consisting of shape descriptors, first-order features, and second-order texture features. Overall, 26.6% and 26.2% of total features demonstrated less than 5% difference between 3D and RG protocols for CT and PET, respectively. Between 10 RG phases in PET, 53.4% of features demonstrated percent differences less than 5%. The features with least variability for PET were sphericity, spherical disproportion, entropy (first and second order), sum entropy, information measure of correlation 2, Short Run Emphasis (SRE), Long Run Emphasis (LRE), and Run Percentage (RPC); and those for CT were minimum intensity, mean intensity, Root Mean Square (RMS), Short Run Emphasis (SRE), and RPC. Quantitative analysis using a 3D acquisition versus RG acquisition (to reduce the effects of motion) provided notably different image feature values. This study suggests that the variability between 3D and RG features is mainly due to the impact of respiratory motion.

Sensitivity of Image Features to Noise in Conventional and Respiratory-Gated PET/CT Images of Lung Cancer: Uncorrelated Noise Effects.

The effect of noise on image features has yet to be studied in depth. Our objective was to explore how significantly image features are affected by the addition of uncorrelated noise to an image. The signal-to-noise ratio and noise power spectrum were calculated for a positron emission tomography/computed tomography scanner using a Ge-68 phantom. The conventional and respiratory-gated positron emission tomography/computed tomography images of 31 patients with lung cancer were retrospectively examined. Multiple sets of noise images were created for each original image by adding Gaussian noise of varying standard deviation equal to 2.5%, 4.0%, and 6.0% of the maximum intensity for positron emission tomography images and 10, 20, 50, 80, and 120 Hounsfield units for computed tomography images. Image features were extracted from all images, and percentage differences between the original image and the noise image feature values were calculated. These features were then categorized according to the noise sensitivity. The contour-dependent shape descriptors averaged below 4% difference in positron emission tomography and below 13% difference in computed tomography between noise and original images. Gray level size zone matrix features were the most sensitive to uncorrelated noise exhibiting average differences >200% for conventional and respiratory-gated images in computed tomography and 90% in positron emission tomography. Image feature differences increased as the noise level increased for shape, intensity, and gray-level co-occurrence matrix features in positron emission tomography and for gray-level co-occurrence matrix and gray-level size zone matrix features in conventional computed tomography. Investigators should be aware of the noise effects on image features.

Fiducial markers coupled with 3D PET/CT offer more accurate radiation treatment delivery for locally advanced esophageal cancer

Background and aims  The role of three-dimensional positron emission tomography/computed tomography (3 D PET/CT) in esophageal tumors that move with respiration and have potential for significant mucosal inflammation is unclear. The aim of this study was to determine the correlation between gross tumor volumes derived from 3 D PET/CT and endoscopically placed fiducial markers. Methods  This was a retrospective, IRB approved analysis of 40 patients with esophageal cancer with fiducials implanted and PET/CT. The centroid of each fiducial was identified on PET/CT images. Distance between tumor volume and fiducials was measured using axial slices. Image features were extracted and tested for pathologic response predictability. Results  The median adaptively calculated threshold value of the standardized uptake value (SUV) to define the metabolic tumor volume (MTV) border was 2.50, which corresponded to a median 23 % of the maximum SUV. The median distance between the inferior fiducial centroid and MTV was – 0.60 cm (– 3.9 to 2.7 cm). The median distance between the superior fiducial centroid and MTV was 1.25 cm (– 4.2 to 6.9 cm). There was no correlation between MTV-to-fiducial distances greater than 2 cm and the gastroenterologist who performed the fiducial implantation. Eccentricity demonstrated statistically significant correlations with pathologic response. Conclusions  There was a stronger correlation between inferior fiducial location and MTV border compared to the superior extent. The etiology of the discordance superiorly is unclear, potentially representing benign secondary esophagitis, presence of malignant nodes, inflammation caused by technical aspects of the fiducial placement itself, or potential submucosal disease. Given the concordance inferiorly and the ability to more precisely set up the patient with daily image guidance matching to fiducials, it may be possible to minimize the planning tumor volume (PTV) margin in select patients, thereby, limiting dose to normal structures.

SU‐E‐QI‐16: Reproducibility of Computed Tomography Quantitative Structural Features Using the FDA Thoracic Phantom Image Database

PURPOSE To assess the reproducibility of quantitative structural features using images from the computed tomography thoracic FDA phantom database under different scanning conditions. METHODS Development of quantitative image features to describe lesion shape and size, beyond conventional RECIST measures, is an evolving area of research in need of benchmarking standards. Gavrielides et al. (2010) scanned a FDA-developed thoracic phantom with nodules of various Hounsfield units (HU) values, shapes and sizes close to vascular structures using several scanners and varying scanning conditions/parameters; these images are in the public domain. We tested six structural features, namely, Convexity, Perimeter, Major Axis, Minor Axis, Extent Mean and Eccentricity, to characterize lung nodules. Convexity measures lesion irregularity referenced to a convex surface. Previously, we showed it to have prognostic value in lung adenocarcinoma. The above metrics and RECIST measures were evaluated on three spiculated (8mm/-300HU, 12mm/+30HU and 15mm/+30HU) and two non-spiculated (8mm/+100HU and 10mm/+100HU) nodules (from layout 2) imaged at three different mAs values: 25, 100 and 200 mAs; on a Phillips scanner (16-slice Mx8000-IDT; 3mm slice thickness). The nodules were segmented semi-automatically using a commercial software tool; the same HU range was used for all nodules. RESULTS Analysis showed convexity having the lowest maximum coefficient of variation (MCV): 1.1% and 0.6% for spiculated and non-spiculated nodules, respectively, much lower compared to RECIST Major and Minor axes whose MCV were 10.1% and 13.4% for spiculated, and 1.9% and 2.3% for non-spiculated nodules, respectively, across the various mAs. MCVs were consistently larger for speculated nodules. In general, the dependence of structural features on mAs (noise) was low. CONCLUSION The FDA phantom CT database may be used for benchmarking of structural features for various scanners and scanning conditions; we used only a small fraction of available data. Our feature convexity outperformed other structural features including RECIST measures.

MO-DE-207B-04: Impact of Reconstruction Field of View On Radiomics Features in Computed Tomography (CT) Using a Texture Phantom.

PURPOSE To investigate the impact of reconstruction Field of View on Radiomics features in computed tomography (CT) using a texture phantom. METHODS A rectangular Credence Cartridge Radiomics (CCR) phantom, composed of 10 different cartridges, was scanned on four different CT scanners from two manufacturers. A pre-defined scanning protocol was adopted for consistency. The slice thickness and reconstruction interval of 1.5 mm was used on all scanners. The reconstruction FOV was varied to result a voxel size ranging from 0.38 to 0.98 mm. A spherical region of interest (ROI) was contoured on the shredded rubber cartridge from CCR phantom CT scans. Ninety three Radiomics features were extracted from ROI using an in-house program. These include 10 shape, 22 intensity, 26 GLCM, 11 GLZSM, 11 RLM, 5 NGTDM and 8 fractal dimensional features. To evaluate the Interscanner variability across three scanners, a coefficient of variation (COV) was calculated for each feature group. Each group was further classified according to the COV by calculating the percentage of features in each of the following categories: COV≤ 5%, between 5 and 10% and ≥ 10%. RESULTS Shape features were the most robust, as expected, because of the spherical contouring of ROI. Intensity features were the second most robust with 54.5 to 64% of features with COV < 5%. GLCM features ranged from 31 to 35% for the same category. RLM features were sensitive to specific scanner and 5% variability was 9 to 54%. Almost all GLZM and NGTDM features showed COV ≥10% among the scanners. The dependence of fractal dimensions features on FOV was not consistent across different scanners. CONCLUSION We concluded that reconstruction FOV greatly influence Radiomics features. The GLZSM and NGTDM are highly sensitive to FOV. funded in part by Grant NIH/NCI R01CA190105-01.

TU-F-CAMPUS-J-05: Effect of Uncorrelated Noise Texture On Computed Tomography Quantitative Image Features

Purpose: To investigate the relationship between quantitative image features (i.e. radiomics) and statistical fluctuations (i.e. electronic noise) in clinical Computed Tomography (CT) using the standardized American College of Radiology (ACR) CT accreditation phantom and patient images. Methods: Three levels of uncorrelated Gaussian noise were added to CT images of phantom and patients (20) acquired in static mode and respiratory tracking mode. We calculated the noise-power spectrum (NPS) of the original CT images of the phantom, and of the phantom images with added Gaussian noise with means of 50, 80, and 120 HU. Concurrently, on patient images (original and noise-added images), image features were calculated: 14 shape, 19 intensity (1st order statistics from intensity volume histograms), 18 GLCM features (2nd order statistics from grey level co-occurrence matrices) and 11 RLM features (2nd order statistics from run-length matrices). These features provide the underlying structural information of the images. GLCM (size 128x128) was calculated with a step size of 1 voxel in 13 directions and averaged. RLM feature calculation was performed in 13 directions with grey levels binning into 128 levels. Results: Adding the electronic noise to the images modified the quality of the NPS, shifting the noise from mostly correlated to mostly uncorrelated voxels. The dramatic increase in noise texture did not affect image structure/contours significantly for patient images. However, it did affect the image features and textures significantly as demonstrated by GLCM differences. Conclusion: Image features are sensitive to acquisition factors (simulated by adding uncorrelated Gaussian noise). We speculate that image features will be more difficult to detect in the presence of electronic noise (an uncorrelated noise contributor) or, for that matter, any other highly correlated image noise. This work focuses on the effect of electronic, uncorrelated, noise and future work shall examine the influence of changes in quantum noise on the features. J. Oliver was supported by NSF FGLSAMP BD award HRD #1139850 and the McKnight Doctoral Fellowship.

SU-E-QI-17: Dependence of 3D/4D PET Quantitative Image Features On Noise

PURPOSE Quantitative imaging is a fast evolving discipline where a large number of features are extracted from images; i.e., radiomics. Some features have been shown to have diagnostic, prognostic and predictive value. However, they are sensitive to acquisition and processing factors; e.g., noise. In this study noise was added to positron emission tomography (PET) images to determine how features were affected by noise. METHODS Three levels of Gaussian noise were added to 8 lung cancer patients PET images acquired in 3D mode (static) and using respiratory tracking (4D); for the latter images from one of 10 phases were used. A total of 62 features: 14 shape, 19 intensity (1stO), 18 GLCM textures (2ndO; from grey level co-occurrence matrices) and 11 RLM textures (2ndO; from run-length matrices) features were extracted from segmented tumors. Dimensions of GLCM were 256×256, calculated using 3D images with a step size of 1 voxel in 13 directions. Grey levels were binned into 256 levels for RLM and features were calculated in all 13 directions. RESULTS Feature variation generally increased with noise. Shape features were the most stable while RLM were the most unstable. Intensity and GLCM features performed well; the latter being more robust. The most stable 1stO features were compactness, maximum and minimum length, standard deviation, root-mean-squared, I30, V10-V90, and entropy. The most stable 2ndO features were entropy, sum-average, sum-entropy, difference-average, difference-variance, difference-entropy, information-correlation-2, short-run-emphasis, long-run-emphasis, and run-percentage. In general, features computed from images from one of the phases of 4D scans were more stable than from 3D scans. CONCLUSION This study shows the need to characterize image features carefully before they are used in research and medical applications. It also shows that the performance of features, and thereby feature selection, may be assessed in part by noise analysis.

A Method to Determine the Optimal Number of Bins in 4D PET

of axillary and extra-axillary metastases identified by FDG PET/CT in patients scheduled for neoadjuvant chemotherapy, and how often this information could change post-operative radiation planning. Materials/Methods: We performed a retrospective analysis of 38 patients with breast cancer scheduled for neoadjuvant chemotherapy between January 2011 and July 2012. 10 patients were clinical stage II, 26 clinical stage III, 2 clinical stage IV. All patients had a FDG PET/CT within 1 month of diagnosis. 28/32 patients had pathologic confirmation of ipsilateral axillary lymph nodes. We identified the incidence of positive axillary lymph nodes and extra axillary metastases, correlated this with stage, and identified how often this could change radiation planning. Results: Axillary lymph nodes were positive in 32/38 patients (84.2%); 5/ 10 (50%) stage II, 25/26 (96.2%) stage III, 2/2 (100%) stage IV. 28/32 (87.5%) of patients with PET positive axillary lymph nodes had pathologic confirmation. 16/38 patients had extra-axillary metastases. These were identified in 14/26 stage III patients (53.8%) and 2/2 stage IV patients (100%). Sites of extra-axillary PET positive metastases were: subpectoral 11/38 (28.9%), internal mammary chain (IMC) 6/38 (15.8%), supraclavicular 2/38 (5.3%), subclavian 1/38 (2.6%), mediastinal lymph node 1/38 (2.6%), and pulmonary nodule 1/38 (2.6%). One patient with positive IMC nodes did not have positive axillary nodes. In all other cases (15/16) patients with extra axillary metastases had axillary metastases. Metastases to subpectoral, IMC, supraclavicular, and subclavian lymph nodes could potentially require modification of post-operative radiation therapy fields. (Total 20/38, 52.6%) Conclusions: FDG PET/CT detected positive axillary lymph nodes in 84.2% of breast cancer patients scheduled for neoadjuvant chemotherapy; in 50% of Stage II patients, 96.2% of stage III patients and 100% of Stage IV patients. Extra-axillary metastases were identified in 42.1% of patients, 53.8% of stage III patients and 100 % of stage IV patients. In 52.6 % of patients, non-axillary regional metastases were identified that could potentially change radiation treatment plans. In clinical stage III and limited stage IV disease, FDG/PET CT could contribute to modified radiation treatment planning.

TH‐A‐137‐07: Local Control Differences for SBRT Lung Patients Planned with Pencil Beam Vs. Collapsed Cone Convolution Algorithms

PURPOSE There are significant differences between pencil beam and convolution algorithms when used in the thorax. However, these differences have not been previously directly correlated with lung SBRT outcomes. METHODS Data from 237 patients treated with SBRT for NSCLC or metastatic disease from other sites were retrospectively analyzed. Of those, 138 were planned using Brainlab treatment planning software (TPS) with the pencil beam (PB) algorithm and treated on Novalis with fiducial marker-based, dual planar radiograph image guidance. Ninety-nine were planned using Pinnacle TPS with collapsed cone convolution (CCC) algorithm (considered a more accurate model) and treated on Varian linacs with CBCT image guidance. Target delineation/expansion guidelines, nominal prescription values, and dose restrictions were identical between the two groups. Median follow-up was 28 and 17 months for the PB and CCC groups, respectively. Kaplan-Meier survival curves were used to determine the statistical differences in local/marginal control between the two groups. Forty-nine plans done in Brainlab (24 with and 25 without local recurrence) were reproduced in Pinnacle with the same MUs. D99 (PTV, GITV), V90 (PTV, GITV), D95 (PTV) and the isocenter dose (D_ISO) were compared. RESULTS Twenty-seven patients (19.6%) planned with the PB and 5 patients (5.1%) planned with the CCC algorithms developed local recurrence. Mantel-Cox and Wilcoxon tests show significant differences between the two groups (P< 0.02). Differences between the two algorithms for target coverage were: D99_GITV=13.9%, D99_PTV=20.8%, V90_GITV=13.7%, V90_PTV=37.5%, D95_PTV=18.9% and D_ISO=6%. CONCLUSION The outcome differences between the patients planned with the PB vs CCC algorithms were statistically significant. Even after controlling for potential variables between the two groups, there appears to be a loss of local control when the delivered dose is about 20% lower than the prescription. Conversely, potential benefit with dose escalation cannot be excluded.