Sébastien Vauclin

Realistic and Efficient Modeling of Radiotracer Heterogeneity in Monte Carlo Simulations of PET Images With Tumors

Monte Carlo simulations are extensively used in PET to evaluate the accuracy with which PET images can yield reliable estimates of parameters of interest. For such applications, the simulated images should be as realistic as possible so that conclusions can be extrapolated to clinical PET images. In this work, we describe a method for introducing realistic modeling of radiotracer heterogeneity into Monte Carlo simulations of patient PET scans. The modeling of the complex physiological activity distribution in healthy regions is directly based on real patient PET/CT images, and realistic tumor shapes can be included into these regions. This method represents a competitive alternative to the use of complex anthropomorphic phantoms such as the XCAT, that require a fixed activity per structure. The method is extended to the simulation of serial PET scans with tumor changes, as acquired in the context of therapy monitoring, and this extension is validated using a patient study. Using the proposed method, very realistic patient PET images can be produced for evaluation purposes.In addition, a strategy to efficiently simulate many sets of pathological cases, based on a unique background physiological activity distribution, is described and carefully assessed using a numerical phantom. The background activity is simulated only once, while tumors are simulated separately. The data are then recombined in a specific way so that the final image has the same properties as images produced by simulating pathological and tumor activities at the same time.

Predictive value of initial FDG-PET features for treatment response and survival in esophageal cancer patients treated with chemo-radiation therapy using a random forest classifier

Purpose In oncology, texture features extracted from positron emission tomography with 18-fluorodeoxyglucose images (FDG-PET) are of increasing interest for predictive and prognostic studies, leading to several tens of features per tumor. To select the best features, the use of a random forest (RF) classifier was investigated. Methods Sixty-five patients with an esophageal cancer treated with a combined chemo-radiation therapy were retrospectively included. All patients underwent a pretreatment whole-body FDG-PET. The patients were followed for 3 years after the end of the treatment. The response assessment was performed 1 month after the end of the therapy. Patients were classified as complete responders and non-complete responders. Sixty-one features were extracted from medical records and PET images. First, Spearman’s analysis was performed to eliminate correlated features. Then, the best predictive and prognostic subsets of features were selected using a RF algorithm. These results were compared to those obtained by a Mann-Whitney U test (predictive study) and a univariate Kaplan-Meier analysis (prognostic study). Results Among the 61 initial features, 28 were not correlated. From these 28 features, the best subset of complementary features found using the RF classifier to predict response was composed of 2 features: metabolic tumor volume (MTV) and homogeneity from the co-occurrence matrix. The corresponding predictive value (AUC = 0.836 ± 0.105, Se = 82 ± 9%, Sp = 91 ± 12%) was higher than the best predictive results found using the Mann-Whitney test: busyness from the gray level difference matrix (P < 0.0001, AUC = 0.810, Se = 66%, Sp = 88%). The best prognostic subset found using RF was composed of 3 features: MTV and 2 clinical features (WHO status and nutritional risk index) (AUC = 0.822 ± 0.059, Se = 79 ± 9%, Sp = 95 ± 6%), while no feature was significantly prognostic according to the Kaplan-Meier analysis. Conclusions The RF classifier can improve predictive and prognostic values compared to the Mann-Whitney U test and the univariate Kaplan-Meier survival analysis when applied to several tens of features in a limited patient database.

Monte Carlo simulations of respiratory gated 18 F-FDG PET for the assessment of volume measurement methods

In PET/CT thoracic imaging, respiratory motion has been reported as a limiting factor reducing image quality and biasing lesion volume measurement. One solution consists in performing respiratory gated PET acquisitions. The aim of this study was to evaluate the impact of respiratory gating on Monte-Carlo realistic PET data, simulated using the 4D-NCAT numerical phantom on the GATE platform. To obtain reconstructed images as close as possible to those obtained in clinical conditions, a particular attention was paid to apply the same type of reconstruction and correction processes on the simulated data as on real clinical ones. The whole set of simulations required a CPU time of 140 000 h generating 1.5 To of data, including simulations of 147 respiratory gated and 49 ungated thoracic exams. Comparison of the displacement volume (DV) measurements using conventional PET acquisitions versus respiratory gated acquisitions was performed, using an automatic iterative segmentation method and a fixed 40% threshold. The segmentation of gated and ungated frames using the 40% fixed threshold needed time consuming initial manual exclusion of noisy structures and so not considered as an automatic method. This step was not necessary when the automatic iterative method was used. Accuracy on DV measurement using the automatic approach was largely improved on gated compared to ungated images. This improved accuracy might have a significant impact when patient treatment is performed using ungated external radiotherapy.

Reproducibility of the adaptive thresholding calibration procedure for the delineation of 18F-FDG-PET-positive lesions.

ObjectiveThe aim of the study was to evaluate the robustness of the calibration procedure against the counting statistics and lesion volumes when using an adaptive thresholding method for the delineation of 2-[18F]fluoro-2-deoxyglucose (18F-FDG)-PET-positive tissue. Materials and methodsThree data sets obtained from physical and simulated images of a phantom containing hot spheres of known volume and contrast were used to study the robustness of the calibration procedure against the counting statistics and range of volumes and contrasts for a given PET model. The mathematical expression of the adaptive thresholding method used corresponds to a linear relationship between the optimal threshold value and the inverse of the local contrast. Robustness was evaluated by testing whether the slopes and intercepts of the linear expression found under two experimental conditions were significantly different (P<0.05). ResultsIt was found that the calibration step was not sensitive to the PET device for the studied PET model, nor to the counting statistics for a signal-to-noise ratio higher than 5.7. No statistical difference was found in the calibration step when using a wide range of volumes (0.2–200 ml) and contrasts (2.0–20.6) or more restricted ones (0.43–97.3 ml and 2.0–7.7, respectively). Therefore, a calibration procedure using limited experimental conditions can be applied to a wider range of volumes and contrasts. ConclusionThese results show that the manufacturer could propose simulated or experimental raw data corresponding to a given PET model with high counting statistics, allowing each clinical center to reconstruct calibration images according to the algorithm parameters used in the clinic.

Retrospective Voxel-Based Dosimetry for Assessing the Ability of the Body-Surface-Area Model to Predict Delivered Dose and Radioembolization Outcome

The aim of this study was to quantitatively evaluate the ability of the body-surface-area (BSA) model to predict tumor-absorbed dose and treatment outcome through retrospective voxel-based dosimetry. Methods: Data from 35 hepatocellular carcinoma patients with a total of 42 90Y-resin microsphere radioembolization treatments were included. Injected activity was planned with the BSA model. Voxel dosimetry based on 99mTc-labeled macroaggregated albumin SPECT and 90Y-microsphere PET was retrospectively performed using a dedicated treatment planning system. Average dose and dose–volume histograms (DVHs) of the anatomically defined tumors were analyzed. The selected dose metrics extracted from DVHs were minimum dose to 50% and 70% of the tumor volume and percentage of the volume receiving at least 120 Gy. Treatment response was evaluated 6 mo after therapy according to the criteria of the European Association for the Study of the Liver. Results: Six-month response was evaluated in 26 treatments: 14 were considered to produce an objective response and 12 a nonresponse. Retrospective evaluation of 90Y-microsphere PET–based dosimetry showed a large interpatient variability with a median average absorbed dose of 60 Gy to the tumor. In 62% (26/42) of the cases, tumor, nontumoral liver, and lung doses would have complied with the recommended thresholds if the injected activity calculated by the BSA method had been increased. Average doses, minimum dose to 50% and 70% of the tumor volume, and percentage of the volume receiving at least 120 Gy were significantly higher in cases of objective response than in nonresponse. Conclusion: In our population, average tumor-absorbed dose and DVH metrics were associated with tumor response. However, the activity calculated by the BSA method could have been increased to reach the recommended tumor dose threshold. Tumor uptake, target and nontarget volumes, and dose distribution heterogeneity should be considered for activity planning.

Multidimensional B-spline parameterization of the detection probability of the PET scanner Biograph 16 using GATE

GATE (Geant4 Application for Emission Tomography) is a widespread, well validated and very versatile application for Monte Carlo simulations in emission tomography. It allows very detailed simulations of positron emission tomography (PET) or single photon emission computed tomography (SPECT) scanners using the general purpose Monte Carlo code Geant4. We present a method to accelerate GATE simulations when on the one hand the detection probability in the different crystals needs to be computed accurately taking into account the complex geometry of the detection system and on the other hand information on photon statistics is not required. The simulation is split into two parts. Inside the scanner (phantom or voxelized patient) the particles are tracked conventionally. When a particle leaves the phantom, the particle is transported to a virtual boundary that replaces the detection system in the simulation and the detection probability is calculated analytically by evaluating a multi-dimensional B-spline function taking into account the state of the photon on this virual border (position, incident angles, energy). The B-spline function itself is calculated by performing a conventional Monte Carlo simulation. This ¿pre¿-simulation needs to be performed only once for each scanner type. The presented method was tested by simulating the PET scanner Biograph HiREZ and using a realistic voxelized attenuation phantom. The results show good agreement between conventional GATE simulations and simulations using the B-spline. The simulations could be accelerated by a factor of around 23 yielding the same statistical accuracy.

Clinical comparison of HiRez versus non-HiRez LSO crystal sampling for lesion detection and SUV quantification

The aim of this work was to evaluate the clinical impact the new HiRez crystal arrangement (Biograph PET scanner) on lesions detection and quantification. Twelve non-HiRez patient sinograms were generated from the original HiRez sinograms. Data conversion (smoothing) from HiRez to non-HiRez geometry was based on real acquisitions and Monte-Carlo simulations of point sources in air for the HiRez and non-HiRez models. The averaged tumor volume reduction on the HiRez images compared to the non-HiRez images for the 41 lesions considered was -13.9 plusmn 19.4% (p < 0.01, paired t-test). Although significant SUVs modifications were found (Max SUVs were statistically higher in HiRez images than in non-HiRez images: 4.2 plusmn 3.1 vs 4.0 plusmn 2.9 respectively; p < 0.001), no difference was observed in terms of lesion detection in our small series, possibly due to lesion characteristics (isolated lesions, high tissue-to-background ratio).