Milan Grkovski

Pharmacokinetic Analysis of Dynamic 18 F-Fluoromisonidazole PET Data in Non–Small Cell Lung Cancer

Hypoxic tumors exhibit increased resistance to radiation, chemical, and immune therapies. 18F-fluoromisonidazole (18F-FMISO) PET is a noninvasive, quantitative imaging technique used to evaluate the magnitude and spatial distribution of tumor hypoxia. In this study, pharmacokinetic analysis (PKA) of 18F-FMISO dynamic PET extended to 3 h after injection is reported for the first time, to our knowledge, in stage III–IV non–small cell lung cancer (NSCLC) patients. Methods: Sixteen patients diagnosed with NSCLC underwent 2 PET/CT scans (1–3 d apart) before radiation therapy: a 3-min static 18F-FDG and a dynamic 18F-FMISO scan lasting 168 ± 15 min. The latter data were acquired in 3 serial PET/CT dynamic imaging sessions, registered with each other and analyzed using pharmacokinetic modeling software. PKA was performed using a 2-tissue, 3-compartment irreversible model, and kinetic parameters were estimated for the volumes of interest determined using coregistered 18F-FDG images for both the volume of interest–averaged and the voxelwise time–activity curves for each patient’s lesions, normal lung, and muscle. Results: We derived average values of 18F-FMISO kinetic parameters for NSCLC lesions as well as for normal lung and muscle. We also investigated the correlation between the trapping rate (k3) and delivery rate (K1), influx rate (Ki) constants, and tissue-to-blood activity concentration ratios (TBRs) for all tissues. Lesions had trapping rates 1.6 times larger, on average, than those of normal lung and 4.4 times larger than those in muscle. Additionally, for almost all cases, k3 and Ki had a significant strong correlation for all tissue types. The TBR–k3 correlation was less straightforward, showing a moderate to strong correlation for only 41% of lesions. Finally, K1–k3 voxelwise correlations for tumors were varied, but negative for 76% of lesions, globally exhibiting a weak inverse relationship (average R = −0.23 ± 0.39). However, both normal tissue types exhibited significant positive correlations for more than 60% of patients, with 41% having moderate to strong correlations (R > 0.5). Conclusion: All lesions showed distinct 18F-FMISO uptake. Variable 18F-FMISO delivery was observed across lesions, as indicated by the variable values of the kinetic rate constant K1. Except for 3 cases, some degree of hypoxia was apparent in all lesions based on their nonzero k3 values.

Feasibility of 18F-Fluoromisonidazole Kinetic Modeling in Head and Neck Cancer Using Shortened Acquisition Times

18F-fluoromisonidazole dynamic PET (dPET) is used to identify tumor hypoxia noninvasively. Its routine clinical implementation, however, has been hampered by the long acquisition times required. We investigated the feasibility of kinetic modeling using shortened acquisition times in 18F-fluoromisonidazole dPET, with the goal of expediting the clinical implementation of 18F-fluoromisonidazole dPET protocols. Methods: Six patients with squamous cell carcinoma of the head and neck and 10 HT29 colorectal carcinoma–bearing nude rats were studied. In addition to an 18F-FDG PET scan, each patient underwent a 45-min 18F-fluoromisonidazole dPET scan, followed by 10-min acquisitions at 96 ± 4 and 163 ± 17 min after injection. Ninety-minute 18F-fluoromisonidazole dPET scans were acquired in animals. Intratumor voxels were classified into 4 clusters based on their kinetic behavior using k-means clustering. Kinetic modeling was performed using the foregoing full datasets (FD) and repeated for each of 2 shortened datasets corresponding to the first approximately 100 min (SD1; patients only) or the first 45 min (SD2) of dPET data. The kinetic rate constants (KRCs) as calculated with a 2-compartment model for both SD1 and SD2 were compared with those derived from FD by correlation (Pearson), regression (Passing–Bablok), deviation (Bland–Altman), and classification (area-under-the-receiver-operating characteristic curve) analyses. Simulations were performed to assess uncertainties due to statistical noise. Results: Strong correlation (r ≥ 0.75, P < 0.001) existed between all KRCs deduced from both SD1 and SD2, and from FD. Significant differences between KRCs were found only for FD-SD2 correlations in patient studies. K1 and k3 were reproducible to within approximately 6% and approximately 30% (FD-SD1; patients) and approximately 4% and approximately 75% (FD-SD2; animals). Area-under-the-receiver-operating characteristic curve values for classification of patient clusters as hypoxic, using a tumor-to-blood ratio greater than 1.2, were 0.91 (SD1) and 0.86 (SD2). The percentage SD in estimating K1 and k3 from 45-min shortened datasets due to noise was less than 1% and between 2% and 12%, respectively. Conclusion: Using single-session 45-min shortened 18F-fluoromisonidazole dPET datasets appears to be adequate for the identification of intratumor regions of hypoxia. However, k3 was significantly overestimated in the clinical cohort. Further studies are necessary to evaluate the clinical significance of differences between the results as calculated from full and shortened datasets.

Multiparametric imaging of tumor hypoxia and perfusion with 18F-FMISO dynamic PET in head and neck cancer

Tumor hypoxia and perfusion are independent prognostic indicators of patient outcome. We developed the methodology for and investigated the utility of multiparametric imaging of tumor hypoxia and perfusion with 18F-fluoromisonidazole (18F-FMISO) dynamic PET (dPET) in head and neck cancer. Methods: One hundred twenty head and neck cancer patients underwent 0- to 30-min 18F-FMISO dPET in a customized immobilization mask, followed by 10-min static acquisitions starting at 93 ± 6 and 160 ± 13 min after injection. A total of 248 lesions (≥2 cm3) were analyzed. Voxelwise pharmacokinetic modeling was conducted using an irreversible 1-plasma 2-tissue-compartment model to calculate surrogate biomarkers of tumor hypoxia (k3), perfusion (K1), and 18F-FMISO distribution volume. The analysis was repeated with truncated dPET datasets. Results: Substantial inter- and intratumor heterogeneity was observed for all investigated metrics. Equilibration between the blood and unbound 18F-FMISO was rapid in all tumors. 18F-FMISO distribution volume deviated from the expected value of unity, causing discrepancy between k3 maps and total 18F-FMISO uptake and reducing the dynamic range of total 18F-FMISO uptake for quantifying the degree of hypoxia. Both positive and negative trends between hypoxia and perfusion were observed in individual lesions. All investigated metrics were reproducible when calculated from a truncated 20-min dataset. Conclusion: 18F-FMISO dPET provides the data necessary to generate parametric maps of tumor hypoxia, perfusion, and radiotracer distribution volume. These data clarify the ambiguity in interpreting 18F-FMISO uptake and improve the characterization of lesions. We show total acquisition times can be reduced to 20 min, facilitating the translation of 18F-FMISO dPET into the clinic.

Multi‐site quality and variability analysis of 3D FDG PET segmentations based on phantom and clinical image data

Purpose: Radiomics utilizes a large number of image‐derived features for quantifying tumor characteristics that can in turn be correlated with response and prognosis. Unfortunately, extraction and analysis of such image‐based features is subject to measurement variability and bias. The challenge for radiomics is particularly acute in Positron Emission Tomography (PET) where limited resolution, a high noise component related to the limited stochastic nature of the raw data, and the wide variety of reconstruction options confound quantitative feature metrics. Extracted feature quality is also affected by tumor segmentation methods used to define regions over which to calculate features, making it challenging to produce consistent radiomics analysis results across multiple institutions that use different segmentation algorithms in their PET image analysis. Understanding each element contributing to these inconsistencies in quantitative image feature and metric generation is paramount for ultimate utilization of these methods in multi‐institutional trials and clinical oncology decision making. Methods: To assess segmentation quality and consistency at the multi‐institutional level, we conducted a study of seven institutional members of the National Cancer Institute Quantitative Imaging Network. For the study, members were asked to segment a common set of phantom PET scans acquired over a range of imaging conditions as well as a second set of head and neck cancer (HNC) PET scans. Segmentations were generated at each institution using their preferred approach. In addition, participants were asked to repeat segmentations with a time interval between initial and repeat segmentation. This procedure resulted in overall 806 phantom insert and 641 lesion segmentations. Subsequently, the volume was computed from the segmentations and compared to the corresponding reference volume by means of statistical analysis. Results: On the two test sets (phantom and HNC PET scans), the performance of the seven segmentation approaches was as follows. On the phantom test set, the mean relative volume errors ranged from 29.9 to 87.8% of the ground truth reference volumes, and the repeat difference for each institution ranged between −36.4 to 39.9%. On the HNC test set, the mean relative volume error ranged between −50.5 to 701.5%, and the repeat difference for each institution ranged between −37.7 to 31.5%. In addition, performance measures per phantom insert/lesion size categories are given in the paper. On phantom data, regression analysis resulted in coefficient of variation (CV) components of 42.5% for scanners, 26.8% for institutional approaches, 21.1% for repeated segmentations, 14.3% for relative contrasts, 5.3% for count statistics (acquisition times), and 0.0% for repeated scans. Analysis showed that the CV components for approaches and repeated segmentations were significantly larger on the HNC test set with increases by 112.7% and 102.4%, respectively. Conclusion: Analysis results underline the importance of PET scanner reconstruction harmonization and imaging protocol standardization for quantification of lesion volumes. In addition, to enable a distributed multi‐site analysis of FDG PET images, harmonization of analysis approaches and operator training in combination with highly automated segmentation methods seems to be advisable. Future work will focus on quantifying the impact of segmentation variation on radiomics system performance.

11C-choline pharmacokinetics in recurrent prostate cancer

The aim of this study was to investigate the value of pharmacokinetic modeling for quantifying 11C-choline uptake in patients with recurrent prostate cancer. Methods: In total, 194 patients with clinically suspected recurrence of prostate cancer underwent 11C-choline dynamic PET over the pelvic region (0–8 min), followed by a 6-min static acquisition at about 25 min after injection. Regions of interest were drawn over sites of disease identified by a radiologist with experience in nuclear medicine. 11C-choline uptake and pharmacokinetics were evaluated by SUV, graphical analysis (Patlak plot; KiP), and 1- and 2-compartment pharmacokinetic models (K1, K1/k2, k3, k4, and the macro parameter KiC). Twenty-four local recurrences, 65 metastatic lymph nodes, 19 osseous metastases, and 60 inflammatory lymph nodes were included in the analysis, which was subsequently repeated for regions of interest placed over the gluteus maximus muscle and adipose tissue as a control. Results: SUVmean and KiP were 3.60 ± 2.16 and 0.28 ± 0.22 min−1 in lesions, compared with 2.11 ± 1.33 and 0.15 ± 0.10 min−1 in muscle and 0.26 ± 0.07 and 0.02 ± 0.01 min−1 in adipose tissue. According to the Akaike information criterion, the 2-compartment irreversible model was most appropriate in 85% of lesions and resulted in a K1 of 0.79 ± 0.98 min−1 (range, 0.11–7.17 min−1), a K1/k2 of 2.92 ± 3.52 (range, 0.31–20.00), a k3 of 0.36 ± 0.30 min−1 (range, 0.00–1.00 min−1) and a KiC of 0.28 ± 0.22 min−1 (range, 0.00–1.33 min−1). The Spearman ρ between SUV and KiP, between SUV and KiC, and between KiP and KiC was 0.94, 0.91, and 0.97, respectively, and that between SUV and K1, between SUV and K1/k2, and between SUV and k3 was 0.70, 0.44, and 0.33, respectively. Malignant lymph nodes exhibited a higher SUV, KiP, and KiC than benign lymph nodes. Conclusion: Although 11C-choline pharmacokinetic modeling has potential to uncouple the contributions of different processes leading to intracellular entrapment of 11C-choline, the high correlation between SUV and both KiP and KiC supports the use of simpler SUV methods to evaluate changes in 11C-choline uptake and metabolism for treatment monitoring.

18F-fluoromisonidazole kinetic modeling for characterization of tumor perfusion and hypoxia in response to antiangiogenic therapy

Multiparametric imaging of tumor perfusion and hypoxia with dynamic 18F-fluoromisonidazole (18F-FMISO) PET may allow for an improved response assessment to antiangiogenic therapies. Cediranib (AZD2171) is a potent inhibitor of tyrosine kinase activity associated with vascular endothelial growth factor receptors 1, 2, and 3, currently in phase II/III clinical trials. Serial dynamic 18F-FMISO PET was performed to investigate changes in tumor biomarkers of perfusion and hypoxia after cediranib treatment. Methods: Twenty-one rats bearing HT29 colorectal xenograft tumors were randomized into a vehicle-treated control group (0.5% methylcellulose daily for 2 d [5 rats] or 7 d [4 rats]) and a cediranib-treated test group (3 mg/kg daily for 2 or 7 d; 6 rats in both groups). All rats were imaged before and after treatment, using a 90-min dynamic PET acquisition after administration of 42.1 ± 3.9 MBq of 18F-FMISO by tail vein injection. Tumor volumes were delineated manually, and the input function was image-derived (abdominal aorta). Kinetic modeling was performed using an irreversible 1-plasma 2-tissue compartmental model to estimate the kinetic rate constants K1, K1/k2, and k3—surrogates for perfusion, 18F-FMISO distribution volume, and hypoxia-mediated entrapment, respectively. Tumor-to-blood ratios (TBRs) were calculated on the last dynamic frame (80–90 min). Tumors were assessed ex vivo by digital autoradiography and immunofluorescence for microscopic visualization of perfusion (pimonidazole) and hypoxia (Hoechst 33342). Results: Cediranib treatment resulted in significant reduction of mean voxelwise 18F-FMISO TBR, K1, and K1/k2 in both the 2-d and the 7-d groups (P < 0.05). The k3 parameter was increased in both groups but reached significance only in the 2-d group. In the vehicle-treated groups, no significant change in TBR, K1, K1/k2, or k3 was observed (P > 0.2). Ex vivo tumor analysis confirmed the presence of hypoxic tumor regions that nevertheless exhibited relatively lower 18F-FMISO uptake. Conclusion: 18F-FMISO kinetic modeling reveals a more detailed response to antiangiogenic treatment than a single static image is able to reveal. The reduced mean K1 reflects a reduction in tumor vascular perfusion, whereas the increased k3 reflects a rise in hypoxia-mediated entrapment of the radiotracer. However, if only late static images are analyzed, the observed reduction in 18F-FMISO uptake after treatment with cediranib may be mistakenly interpreted as a global decrease, rather than an increase, in tumor hypoxia. These findings support the use of 18F-FMISO kinetic modeling to more accurately characterize the response to treatments that have a direct effect on tumor vascularization and perfusion.

Predicting hypoxia status using a combination of contrast-enhanced computed tomography and [ 18 F]-Fluorodeoxyglucose positron emission tomography radiomics features

BACKGROUND AND PURPOSE Hypoxia is a known prognostic factor in head and neck cancer. Hypoxia imaging PET radiotracers such as 18F-FMISO are promising but not widely available. The aim of this study was therefore to design a surrogate for 18F-FMISO TBRmax based on 18F-FDG PET and contrast-enhanced CT radiomics features, and to study its performance in the context of hypoxia-based patient stratification. METHODS 121 lesions from 75 head and neck cancer patients were used in the analysis. Patients received pre-treatment 18F-FDG and 18F-FMISO PET/CT scans. 79 lesions were used to train a cross-validated LASSO regression model based on radiomics features, while the remaining 42 were held out as an internal test subset. RESULTS In the training subset, the highest AUC (0.873±0.008) was obtained from a signature combining CT and 18F-FDG PET features. The best performance on the unseen test subset was also obtained from the combined signature, with an AUC of 0.833, while the model based on the 90th percentile of 18F-FDG uptake had a test AUC of 0.756. CONCLUSION A radiomics signature built from 18F-FDG PET and contrast-enhanced CT features correlates with 18F-FMISO TBRmax in head and neck cancer patients, providing significantly better performance with respect to models based on 18F-FDG PET only. Such a biomarker could potentially be useful to personalize head and neck cancer treatment at centers for which dedicated hypoxia imaging PET radiotracers are unavailable.

Inter-operator variability in compartmental kinetic analysis of 18 F-fluoromisonidazole dynamic PET

PURPOSE To assess the inter-operator variability in compartment analysis (CA) of dynamic-FMISO (dyn-FMISO) PET. METHODS Study-I: Five investigators conducted CA for 23 NSCLC dyn-FMISO tumor time-activity-curves. Study-II: Four operators performed CA for four NSCLC dyn-FMISO datasets. Repeatability of Kinetic-Rate-Constants (KRCs) was assessed. RESULTS Study-I: Strong correlation (ICC > 0.9) and interchangeable results among operators existed for all KRCs. Study-II: Up to 103% variability in tumor segmentation, and weaker ICC in KRCs (ICC-VB = 0.53; ICC-K1 = 0.91; ICC-K1/k2 = 0.25; ICC-k3 = 0.32; ICC-Ki = 0.54) existed. All KRCs were repeatable among the different operators. CONCLUSIONS Inter-operator variability in CA of dyn-FMISO was shown to be within statistical errors.

18 F-fluoromisonidazole predicts evofosfamide uptake in pancreatic tumor model

BackgroundQuantitative imaging can facilitate patient stratification in clinical trials. The hypoxia-activated prodrug evofosfamide recently failed a phase III trial in pancreatic cancer. However, the study did not attempt to select for patients with hypoxic tumors. We tested the ability of 18F-fluoromisonidazole to predict evofosfamide uptake in an orthotopic xenograft model (BxPC3).MethodsTwo forms of evofosfamide were used: (1) labeled on the active moiety (3H) and (2) on the hypoxia targeting nitroimidazole group (14C). Tumor uptake of evofosfamide and 18F-fluoromisonidazole was counted ex vivo. Autoradiography of 14C and 18F coupled with pimonidazole immunohistochemistry revealed the spatial distributions of prodrug, radiotracer, and hypoxia.ResultsThere was significant individual variation in 18F-fluoromisonidazole uptake, and a significant correlation between normalized 18F-fluoromisonidazole and both 3H-labeled and 14C-labeled evofosfamide. 18F-fluoromisonidazole and 14C-evofosfamide both localized in hypoxic regions as identified by pimonidazole.Conclusion18F-fluoromisonidazole predicts evofosfamide uptake in a preclinical pancreatic tumor model.

WE-AB-202-11: Radiobiological Modeling of Tumor Response During Radiotherapy Based On Pre-Treatment Dynamic PET Imaging Data

PURPOSE To evaluate the ability of a multiscale radiobiological model of tumor response to predict mid-treatment hypoxia images, based on pretreatment imaging of perfusion and hypoxia with [18-F]FMISO dynamic PET and glucose metabolism with [18-F]FDG PET. METHODS A mechanistic tumor control probability (TCP) radiobiological model describing the interplay between tumor cell proliferation and hypoxia (Jeong et al., PMB 2013) was extended to account for intra-tumor nutrient heterogeneity, dynamic cell migration due to nutrient gradients, and stromal cells. This extended model was tested on 10 head and neck cancer patients treated with chemoradiotherapy, randomly drawn from a larger MSKCC protocol involving baseline and mid-therapy dynamic PET scans. For each voxel, initial fractions of proliferative and hypoxic tumor cells were obtained by finding an approximate solution to a system of linear equations relating cell fractions to voxel-level FDG uptake, perfusion (FMISO K1 ) and hypoxia (FMISO k3 ). The TCP model then predicted their evolution over time up until the mid treatment scan. Finally, the linear model was reapplied to predict each lesions median hypoxia level (k3 [med,sim]) which in turn was compared to the FMISO k3 [med] measured at mid-therapy. RESULTS The average k3[med] of the tumors in pre-treatment scans was 0.0035 min-1 , with an inter-tumor standard deviation of σ[pre]=0.0034 min-1 . The initial simulated k3 [med,sim] of each tumor agreed with the corresponding measurements within 0.1σ[pre]. In 7 out of 10 lesions, the mid-treatment k3 [med,sim] prediction agreed with the data within 0.3σ[pre]. The remaining cases corresponded to the most extreme relative changes in k3 [med]. CONCLUSION This work presents a method to personalize the prediction of a TCP model using pre-treatment kinetic imaging data, and validates the modeling of radiotherapy response by predicting changes in median hypoxia values during treatment. Variations from predicted response may be a useful biomarker, which should be further explored. Partially supported by NIH grant #1 R01 CA157770-01A1 and a grant from Varian Corporation.