N Jallow

Variability of textural features in FDG PET images due to different acquisition modes and reconstruction parameters

Abstract Background. Characterization of textural features (spatial distributions of image intensity levels) has been considered as a tool for automatic tumor segmentation. The purpose of this work is to study the variability of the textural features in PET images due to different acquisition modes and reconstruction parameters. Material and methods. Twenty patients with solid tumors underwent PET/CT scans on a GE Discovery VCT scanner, 45–60 minutes post-injection of 10 mCi of [18F]FDG. Scans were acquired in both 2D and 3D modes. For each acquisition the raw PET data was reconstructed using five different reconstruction parameters. Lesions were segmented on a default image using the threshold of 40% of maximum SUV. Fifty different texture features were calculated inside the tumors. The range of variations of the features were calculated with respect to the average value. Results. Fifty textural features were classified based on the range of variation in three categories: small, intermediate and large variability. Features with small variability (range ≤ 5%) were entropy-first order, energy, maximal correlation coefficient (second order feature) and low-gray level run emphasis (high-order feature). The features with intermediate variability (10% ≤ range ≤ 25%) were entropy-GLCM, sum entropy, high gray level run emphsis, gray level non-uniformity, small number emphasis, and entropy-NGL. Forty remaining features presented large variations (range > 30%). Conclusion. Textural features such as entropy-first order, energy, maximal correlation coefficient, and low-gray level run emphasis exhibited small variations due to different acquisition modes and reconstruction parameters. Features with low level of variations are better candidates for reproducible tumor segmentation. Even though features such as contrast-NGTD, coarseness, homogeneity, and busyness have been previously used, our data indicated that these features presented large variations, therefore they could not be considered as a good candidates for tumor segmentation.

Heterogeneity in Intratumor Correlations of 18F-FDG, 18F-FLT, and 61Cu-ATSM PET in Canine Sinonasal Tumors

Intratumor heterogeneity in biologic properties and in relationships between various phenotypes may present a challenge for biologically targeted therapies. Understanding the relationships between different phenotypes in individual tumor types could help inform treatment selection. The goal of this study was to characterize spatial correlations of glucose metabolism, proliferation, and hypoxia in 2 histologic types of tumors. Methods: Twenty canine veterinary patients with spontaneously occurring sinonasal tumors (13 carcinomas and 7 sarcomas) were imaged with 18F-FDG, 18F-labeled 3′-deoxy-3′-fluorothymidine (18F-FLT), and 61Cu-labeled diacetyl-bis(N4-methylthiosemicarbazone) (61Cu-ATSM) PET/CT on 3 consecutive days. Precise positioning and immobilization techniques coupled with anesthesia enabled motionless scans with repeatable positioning. Standardized uptake values (SUVs) of gross sarcoma and carcinoma volumes were compared by use of Mann–Whitney U tests. Patient images were rigidly registered together, and intratumor tracer uptake distributions were compared. Voxel-based Spearman correlation coefficients were used to quantify intertracer correlations, and the correlation coefficients of sarcomas and carcinomas were compared. The relative overlap of the highest uptake volumes of the 3 tracers was quantified, and the values were compared for sarcomas and carcinomas. Results: Large degrees of heterogeneity in SUV measures and phenotype correlations were observed. Carcinoma and sarcoma tumors differed significantly in SUV measures, with carcinoma tumors having significantly higher 18F-FDG maximum SUVs than sarcoma tumors (11.1 vs. 5.0; P = 0.01) as well as higher 61Cu-ATSM mean SUVs (2.6 vs. 1.2; P = 0.02). Carcinomas had significantly higher population-averaged Spearman correlation coefficients than sarcomas in comparisons of 18F-FDG and 18F-FLT (0.80 vs. 0.61; P = 0.02), 18F-FLT and 61Cu-ATSM (0.83 vs. 0.38; P < 0.0001), and 18F-FDG and 61Cu-ATSM (0.82 vs. 0.69; P = 0.04). Additionally, the highest uptake volumes of the 3 tracers had significantly greater overlap in carcinomas than in sarcomas. Conclusion: The relationships of glucose metabolism, proliferation, and hypoxia were heterogeneous across different tumors, with carcinomas tending to have high correlations and sarcomas having low correlations. Consequently, canine carcinoma tumors are robust targets for therapies that target a single biologic property, whereas sarcoma tumors may not be well suited for such therapies. Histology-specific PET correlations have far-reaching implications for the robustness of biologic target definition.

Repeatability of Quantitative 18F-NaF PET: A Multicenter Study

18F-NaF, a PET radiotracer of bone turnover, has shown potential as an imaging biomarker for assessing the response of bone metastases to therapy. This study aimed to evaluate the repeatability of 18F-NaF PET–derived SUV imaging metrics in individual bone lesions from patients in a multicenter study. Methods: Thirty-five castration-resistant prostate cancer patients with multiple metastases underwent 2 whole-body (test–retest) 18F-NaF PET/CT scans 3 ± 2 d apart from 1 of 3 imaging sites. A total of 411 bone lesions larger than 1.5 cm3 were automatically segmented using an SUV threshold of 15 g/mL. Two levels of analysis were performed: lesion-level, in which measures were extracted from individual-lesion regions of interest (ROI), and patient-level, in which all lesions within a patient were grouped into a patient ROI for analysis. Uptake was quantified with SUVmax, SUVmean, and SUVtotal. Test–retest repeatability was assessed using Bland–Altman analysis, intraclass correlation coefficient (ICC), coefficient of variation, critical percentage difference, and repeatability coefficient. The 95% limit of agreement (LOA) of the ratio between test and retest measurements was calculated. Results: At the lesion level, the coefficient of variation for SUVmax, SUVmean, and SUVtotal was 14.1%, 6.6%, and 25.5%, respectively. At the patient level, it was slightly smaller: 12.0%, 5.3%, and 18.5%, respectively. ICC was excellent (>0.95) for all SUV metrics. Lesion-level 95% LOA for SUVmax, SUVmean, and SUVtotal was (0.76, 1.32), (0.88, 1.14), and (0.63, 1.71), respectively. Patient-level 95% LOA was slightly narrower, at (0.79, 1.26), (0.89, 1.10), and (0.70, 1.44), respectively. We observed significant differences in the variance and sample mean of lesion-level and patient-level measurements between imaging sites. Conclusion: The repeatability of SUVmax, SUVmean, and SUVtotal for 18F-NaF PET/CT was similar between lesion- and patient-level ROIs. We found significant differences in lesion-level and patient-level distributions between sites. These results can be used to establish 18F-NaF PET–based criteria for assessing treatment response at the lesion and patient levels. 18F-NaF PET demonstrates repeatability levels useful for clinically quantifying the response of bone lesions to therapy.

Diagnostic Reference Levels of CT Radiation Dose in Whole-Body PET/CT

The role of CT in PET/CT imaging includes acquisition techniques for diagnostic, anatomic localization, and attenuation correction purposes. Diagnostic reference levels of the volumetric CT dose index (CTDIvol) are available for dedicated CT procedures on selected body regions, but similar reference levels for whole-body CT used in PET/CT examinations are limited. This work reports CTDIvol values from sites that conduct whole-body oncologic PET/CT examinations and participated in the scanner validation program of the Society of Nuclear Medicine and Molecular Imaging Clinical Trials Network. Methods: From 2010 to 2014, a total of 154 sites submitted CT acquisition parameters used in their clinical 18F-FDG PET/CT oncology protocols. From these parameters, the CTDIvol was estimated using the ImPACT CTDI dosimetry tables. Histograms of CTDIvol values were created for each year, and descriptive statistics, including mean, median, and 75th percentile, were reported. Repeated-measures ANOVA was performed to determine whether significant differences occurred between reporting years. Results: A wide range of technical parameters was reported, most notably in tube current. Between 2010 and 2014, the median CTDIvol ranged from 4.9 to 6.2 mGy and the 75th percentile from 9.7 to 10.2 mGy. There was no significant change in CTDIvol between reporting years (repeated-measures ANOVA, P = 0.985). Conclusion: The 75th percentile CTDIvol reported in this work was 9.8 mGy averaged over all reporting years. These data provide a resource for establishing CTDIvol reference values specific to performing CT in PET/CT whole-body examinations. The wide ranges of CT acquisition parameters reported by sites suggest that CTDIvol reference levels may be beneficial for optimization of CT protocols.

TU‐D‐209‐01: Dosimetry of Diagnostic Work Up Mammography

PURPOSE To investigate patient average glandular dose (AGD) characteristics of diagnostic mammography. METHODS The techniques used to image 14420 patients who received diagnostic work up mammography from October 2008 to December 2014 at one academic hospital were retrospectively collected. The most common diagnostic views and the techniques used for each according to compressed breast thickness were determined. For all techniques, 1st half value layer and air kerma output per tube current-exposure time product were measured; then the incident air kerma for each acquisition was calculated. The values for normalized glandular dose (DgN) were obtained with a validated Monte Carlo simulation of mammographic acquisition. The mono-energetic DgN results were combined according to relative fluence using the TASMICS model to obtain DgN coefficients for each spectrum. The spectral DgN and calculated incident air kerma were used to estimate AGD of patients with breast thickness ranging from 2 to 8 cm. RESULTS The most common views utilized during diagnostic mammography were magnification craniocaudal (24%), magnification mediolateral (19%), spot craniocaudal (28%), and spot mediolateral oblique (24%). The AGD increased with increasing breast thickness for both the magnification and spot views. The AGD for a 5.5 cm thick breast was approximately 6.8 mGy and 2.2 mGy for the magnification and spot views, respectively. The AGD ranged from 3.6 mGy to 6.8 mGy for the magnification views and from 1.0 mGy to 3.1 mGy for spot views. The difference in AGD between the two magnification views or the two spot views was not significant. CONCLUSION These results provide information on breast dose to which screening recalled women are exposed to. In addition to understanding the dose used for common clinical imaging tests, this data could be used when comparing use of mammography for diagnostic workup to other potential modalities, such as breast tomosynthesis and breast CT.

WE‐C‐WAB‐01: Intratumor Correlations of FDG, FLT, and Cu‐ATSM PET in Canine Tumors: Implications for Dose Painting

PURPOSE In dose painting, it is uncertain which of a tumors biological properties should be targeted, and if plans for different tumor histologies are equally sensitive to the choice of biological target. This study characterizes the relationships between three potential biological targets - glucose metabolism, proliferation, and hypoxia - in two different tumor histologies using PET/CT imaging. METHODS Twenty canine patients with sinonasal tumors (7 sarcomas and 13 carcinomas) were imaged using FDG, FLT, and Cu-ATSM PET/CT on three consecutive days. Patients were immobilized and precisely positioned, and resulting images were rigidly registered. Within each tumor volume, voxel SUV distributions from different tracers were compared and inter-tracer correlations were evaluated using voxel-based Spearman correlation coefficients. Correlation coefficients were then Fisher-transformed, and a two-sided t-test was applied to determine if sarcoma and carcinoma populations differed significantly in inter-tracer correlations. SUV measures such as SUVmax, SUVpeak, and SUVmean were also compared between sarcomas and carcinomas using Mann-Whitney U-tests. RESULTS Significant differences in inter-tracer correlations were observed between sarcoma and carcinoma tumors. Population-averaged Spearman correlation coefficients were significantly higher for carcinoma tumors than sarcoma tumors in comparisons of FLT:Cu-ATSM (0.83 vs. 0.38; p<0.0001), FDG:FLT (0.80 vs. 0.61; p=0.02), and FDG:Cu-ATSM (0.82 vs. 0.69; p=0.04). Tracer distributions generally overlapped in carcinomas; in sarcomas, however, different tracers clustered in different tumor regions. Carcinomas also had significantly higher average FDG SUVmax (11.1 vs. 5.0; p=0.01) and higher Cu-ATSM SUVmean (2.6 vs. 1.2; p=0.02) than sarcoma tumors. CONCLUSION Carcinoma tumors, with high spatial correlations between tumor metabolism, proliferation, and hypoxia, are robust targets for therapies that target a single biological property. Sarcomas may not be well-suited for such therapies. Histology-specific robustness in biological target definition has large implications for dose painting strategies, as well as for other biologically targeted therapies.

WE‐C‐BRA‐04: Stability of FLT and Cu‐ATSM PET in Canine Patients during Radiotherapy: Possible Targets for Dose Painting

Purpose: In dose painting, functional imaging is used to identify and target radiotherapy‐resistant regions of the tumor. It is important to know if target locations remain stable during radiotherapy, otherwise dose plans will need to adapt to mid‐therapy changes. This study investigated the spatio‐temporal stability of FLT and Cu‐ATSM PET after a few fractions of radiotherapy.Methods: Nineteen canine patients with nasal tumors (12 carcinoma and 7 sarcoma) were imaged prior to accelerated hypofractionated radiotherapy with [F‐18]FLT PET/CT for proliferation and [Cu‐61]Cu‐ATSM PET/CT for hypoxia. FLT scans were repeated after two fractions (8.4 or 10 Gy) and Cu‐ATSM scans after three fractions (12.6 or 15 Gy). Patients were immobilized with custom bite blocks and vacuum mattresses for reproducible positioning. Mid‐treatment scans were rigidly registered to pre‐treatment scans. Spearman correlation coefficients quantified the voxel‐by‐voxel SUV correlations between pre‐ and mid‐treatment scans. Population and histology averaged correlation coefficients were determined using Fisher transformations. SUVmax, SUVmean, SUVpeak, and SUVtotal were also compared between pre‐ and mid‐treatment scans using paired t‐tests. Results: Pre‐ and mid‐treatment scans were highly correlated, for both FLT and Cu‐ATSM PET distributions. The Fisher‐weighted mean Spearman correlation coefficient was 0.87[range: 0.57–0.95] for Cu‐ATSM and 0.79[0.18–0.90] for FLT. On average, sarcomas had slightly higher correlations (0.88) than carcinomas (0.78). We observed significant reductions (p 0.20) were observed in sarcomas.Conclusion: Spatial distributions of hypoxia and proliferation were stable after a few fractions of radiotherapy. The magnitude of proliferation reduced significantly in both sarcoma and carcinomatumors. The magnitude of hypoxia only reduced significantly in carcinomatumors. With high spatial stability early during radiotherapy, pre‐treatment FLT and Cu‐ATSM PET are potential candidates for dose painting targets.

SU-E-I-176: Impact of Residual Activity and Clock Synchronization on Quantitative Treatment Response Assessment

Purpose: The accuracy of quantitative PET is influenced by many factors whose uncertainties need to be characterized for its reliable use in treatment response assessment. Among the factors are residual activity and lack of synchronization between PET‐scanner and dose‐calibrator clocks which are frequently overlooked. Therefore, we characterized their effects on PET‐ based treatment response assessment. Methods: Fifteen patients receiving targeted molecular therapy underwent whole body [18F]FLT PET/CT scans at multiple time points. Response was calculated as the change in SUV between subsequent scans. Residual‐activity left in syringe after injection effects were assessed by comparing response measures when SUV values were corrected for residual activity (range: 0.07mCi‐0.61mCi) to when they were not. Uncertainty in response was also compared between scans where the synchronization of the PET‐scanner and dose‐calibrator clocks were randomly offset (range: 1min‐60mins) to when they were synchronized. The effect of both factors on PERCIST classification of treatment response was assessed. Results: Residual‐activity and lack of clock synchronization had significant effects on treatment response (p=0.05). 3% change in clock offsets between subsequent scans resulted in 2.5% absolute difference in response. The mean difference in responses due to desynchronized clocks was 18% (maximum=35%). 3% change in residual activity resulted in 0.1% absolute difference in response. Not accounting for residual‐activity caused mean difference of 1% (maximum=6%) in the response. Changes in clock synchronization and not accounting for residual activity resulted in reclassification of the PERCIST response status of over 30% of the patients. Conclusions: Lack of synchronization between PET‐scanner and dose‐ calibrator clocks and not accounting for residual activity caused a difference in response. Administration procedures that minimize residual activity (e.g. flushing syringe with saline) have to be implemented. Also, all clocks used with dose‐calibrator have to be synchronized with PET‐scanner clock periodically. Otherwise patients can be wrongly classified as partial, stable, or progressive disease.

SU‐E‐J‐87: Imaging Biomarkers of Treatment Response: A Roadmap to Validation

Purpose: Quantitative imaging measures are essential for assessment of tumor response to therapy; however, systematic validation is required before imaging biomarkers can be successfully implemented. We performed preliminary validation of the optimal image reconstruction parameters to maximize the accuracy of PET‐based imaging biomarkers. Methods: Optimal PETreconstruction parameters were determined using four spheres of various sizes (10–22mm) containing 168 kBq/cc of [18F]FDG. Spheres were placed inside a NEMA IEC body phantom. Scans were acquired on a GE VCT PET/CT scanner in both 2D and 3D mode. Multiple images were reconstructed using OSEM by varying matrix dimensions, iteration number, and post‐filtration. For each sphere, the recovery coefficient (RC) was used to measure quantitative image accuracy. Robustness of reconstruction parameters to sphere size changes was measured as the coefficient of variation (CV) of the RCs of the individual spheres within an image. Optimal reconstruction parameters were those which resulted in the highest average RC and lowest CV. Results: RC averaged over all four spheres ranged from 0.72 – 0.95 for the image reconstructions investigated. Higher RCs were observed for 2D acquisition mode than 3D (0.91 vs. 0.80), and for 22mm spheres than 10mm spheres (0.99 vs. 0.74). No such trends were observed for CVs. CV values varied by a factor of two across the reconstructions investigated, ranging from 9% to 17%. Image accuracy and robustness to lesion size changes were highest for 2D acquisition, 256×256 matrix, 2 iterations, and 3mm post‐filtration, yielding an RC of 0.92 and CV of ±9%. Conclusions: We successfully optimized image reconstruction parameters to maximize quantitative imaging accuracy, which will increase reproducibility of biomarkers. Future work will include test‐retest imaging of biomarkers in patients to quantify remaining uncertainties. This study represents a critical first step towards validation of imaging biomarkers of treatment response.

TH‐C‐220‐03: Extraction of Radiosensitivity Coefficients from Multiparametric Molecular Imaging Data through Computational Tumor Modeling

Purpose: Patient‐specific differences in tumor radiosensitivity are generally disregarded in radiation therapy (RT), an approach that overlooks subgroups with different therapeutic responses. In this study, we extracted voxel‐based radiosensitivity coefficients from multiparametric molecular imaging data using a computational tumor growth and response model. Methods: Thirteen canine patients with sinonasal tumors underwent pre‐therapy PET/CT imaging of metabolic activity ([18F]FDG), proliferation ([18F]FLT), and hypoxia ([61Cu]Cu‐ATSM) using bite‐block immobilization. Following two RT fractions and two off‐treatment days, proliferative response was assessed via [18F]FLT‐PET/CT. Using a previously developed tumor simulation model, voxel‐based ‘effective’ linear radiosensitivity coefficients (alpha) were determined by simulating treatment response based upon pre‐therapy PET data and iteratively minimizing differences between simulated and imaged proliferative response. Resulting alpha values were investigated for inter‐ and intra‐patient heterogeneity, histology dependence, and spatial correlations with pre‐therapy PET data. Results: Alpha values in carcinomas (mean=0.061/Gy, SD=0.051/Gy) were higher than those found in sarcomas (mean=0.042/Gy, SD=0.024/Gy). Carcinomas were more heterogeneous in response (range=0.554/Gy) than sarcomas (range=0.110/Gy), which correlated with a more heterogeneous pre‐therapy uptake of all three PET imaging surrogate markers. Increased [61Cu]Cu‐ATSM uptake (SUV>4) alone indicated radioresistance, while simultaneous [18F]FLT uptake lowered effective alpha values. Using Akaikes information criterion, five carcinomas were identified to exhibit bimodal or trimodal alpha distributions, while all sarcomas were found to be unimodal. Conclusion: Within this patient population, carcinomas were more radiosensitive and heterogeneous in alpha and PET tracer uptake. Extracted alpha values and their correlations with pre‐therapy PET data differed between tumor histologies. Results suggest that imaged hypoxia might not necessarily indicate increased levels of radioresistance as high proliferation potentially offsets this effect. After correlating early treatment response to late outcome, this model could be used to identify potential biological targets within tumors, stratify candidate dose painting patients, and to transform voxel‐ based information into biological treatment planning objectives.