Optimal respiratory-gated [¹⁸F]FDG PET/CT significantly impacts the quantification of metabolic parameters and their correlation with overall survival in patients with pancreatic ductal adenocarcinoma

In this retrospective explorative study, data from all patients who underwent an [¹⁸F]FDG PET/CT scan at the Radboudumc in Nijmegen for suspected pancreatic malignancy between November 2004 and January 2015 were collected (Table 1). We also included patients who were scanned elsewhere and were referred for further diagnostic evaluation. The available histopathological data acquired by either biopsy or surgery were analyzed and 158 patients with histologically proven pancreatic ductal adenocarcinoma were identified (Additional file 1: Figure S1, CONSORT flowchart). In sixty-nine of these patients, an ORG reconstructed scan was available. We used this group to test the additional value of ORG scanning on the quantification of metabolic parameters. Both non-ORG (according to EANM guidelines) and ORG scans were correlated to OS. Given the retrospective nature of this study and the anonymized handling of data, informed consent was waived by the institutional review board (protocol CMO2018-4420). Scans were anonymized before review and the readers were unaware of the clinical or histopathological data.

All patients fasted for a minimum of 6 h prior to intravenous administration of [¹⁸F]FDG. Subsequently, patients rested for approximately 60 min before imaging.

Whole-body PET images were acquired using a Biograph 40 mCT (Siemens Medical Solutions, Knoxville Tennessee, USA) PET/CT scanner equipped with a dual-slice CT. PET images were reconstructed according to the EANM procedure guidelines for tumor PET imaging [26]. A TrueX algorithm was used with point spread function (PSF) and time of flight (TOF) measurements, using three iterations, 21 subsets, matrix size 200 × 200 (pixel spacing of 4.07 mm), full width half maximum (FWHM) of 3 mm, and using 2 min of PET data. Post-processing was performed using a 3D Gaussian filter kernel, 3.0 mm. For respiratory gating, we used an amplitude-based ORG algorithm, which is integrated in the Syngo 2012A MI.PET/CT software by Siemens, designated HD Chest. The procedures were performed as previously described [18]. In short, the main user input for the ORG algorithm is the percentage duty cycle, which is the percentage of the total acquired true coincidences used for image reconstruction. The ORG algorithm calculates an optimal amplitude range for a given duty cycle, by calculating the amplitude range for different values of the lower limit (L). With each value of L, the upper limit (U) is adjusted to include the specified percentage of the acquired PET data, and an amplitude range (W) is calculated through a simple subtraction (U − L). The optimal amplitude range is defined as the smallest amplitude range obtained and calculated by minimizing W.

The reconstruction settings were kept constant between ORG and non-ORG images; the percentage of duty cycles (total acquired true coincidences used for image reconstruction) was set at 35%, corresponding with 2 min of PET data.

Low-dose CT scans for attenuation corrections were reconstructed using a B19f convolution kernel, slice thickness 5.0 mm, and CT images for anatomical reference were reconstructed using a B31f convolution kernel, slice thickness 3.0 mm.

All [¹⁸F]FDG PET/CT scans were evaluated using the Inveon Research Workspace 4.2 (Preclinical Solutions, Siemens Medical Solutions USA, Knoxville Tennessee, USA). After manual annotation of tumor region, a 40% SUV_max isocontour was used to delineate the metabolic tumor volume. Uptake of [¹⁸F]FDG projecting over adjacent structures, for example, uptake in the duodenum or biliary stent, was manually excluded from the VOIs, unless this was identified as tumor on contrast-enhanced CT (ceCT) or magnetic resonance cholangiopancreatography (MRCP). The first and second-order features were extracted using the PyRadiomics toolbox [27]. For the texture feature extraction, the images were discretized using a fixed number of 255 bins. The texture features were extracted from the co-occurrence (GLCM), gray-level run-length (GLRLM), and gray-level size-zone texture matrices (GLSZM). These matrices were generated by considering 26 connected voxels, 13 directions in three dimensions with a distance of 1 voxel, as described previously [27].

Optimal cutoff values were determined for each individual parameter on both ORG and non-ORG scans, based on Cox proportional hazard model and log-rank testing in Kaplan Meier survival curves. Overall survival of a patient was defined from the date of the PET/CT acquisition until the date of the patient’s death.

Univariate analysis of the data was performed on all 69 non-ORG and ORG scans using the Cox proportional hazard model, after scaling the features using the standard-scaler of Scikit-learn [28]. Additionally, all respiratory-gated scans were compared to their corresponding non-ORG counterpart to test the additional value of respiratory-gating on quantification of metabolic parameters.

Pancreatic lesions can present in a wide range of variations on [¹⁸F]FDG PET/CT, but commonly, ductal adenocarcinomas are FDG avid beyond the level of liver uptake, with corresponding hypodense lesions on ceCT. Smaller lesions can be overlooked as physiological [¹⁸F]FDG accumulation in the duodenum. Small size and low absolute level of [¹⁸F]FDG accumulation renders small lesions particularly prone to partial-volume effects and motion artifacts. Correction for respiratory motion artifacts improved the visual detection of small pancreatic ductal adenocarcinomas in the pancreatic head and ampullary carcinomas (Fig. 1).

Using ORG reconstruction, the visual assessment of intra-tumoral regional differences in [¹⁸F]FDG accumulation changed, revealing regions of absent [¹⁸F]FDG accumulation or multiple foci of increased [¹⁸F]FDG accumulation (Fig. 2).

Next, we compared pairs of standard and ORG reconstructed scans to explore the impact of respiratory motion on the quantification of metabolic parameters. In concordance with previous studies using respiratory gating techniques, the parameters SUV_max, SUV_mean, and SUV_diff showed significantly higher values on ORG scans compared to the corresponding non-ORG scans. The other metabolic parameters, e.g., MTV40%, significantly decreased (Additional file 2: Table S1, Fig. 3).

First-order features that describe intratumoral heterogeneity changed, for example, skewness, kurtosis, standard deviation, and entropy of the histogram increased significantly in the ORG reconstructed scan. Except for the features difference entropy and gray level variance, second-order features also showed a significant increase or decrease in the ORG reconstructed scans, for example, dissimilarity and joint entropy.

As the quantification of metabolic parameters changed considerably using ORG, we investigated whether their correlation with OS was consequently affected. For both standard and ORG reconstructed scans, optimal cutoff values for correlation with OS were defined. Logistic regression analysis demonstrated that the correlation with OS of most metabolic parameters changed considerably, but in an inconsistent pattern (Table 2). Standard PET-derived parameters SUV_max, SUV_mean, and SUV_diff were no longer associated with OS after ORG reconstruction (Table 2, Fig. 4), whereas parameters that describe heterogeneity (sum entropy) became significant, or borderline significant (difference entropy, joint entropy), associated with OS (Table 2, Fig. 5).