Sensitivity of 18F-fluorodihydrotestosterone PET-CT to count statistics and reconstruction protocol in metastatic castration-resistant prostate cancer

Abstract

Objectives: Whole body [¹⁸F]-fluorodihydrotestosterone positron emission tomography ([¹⁸F]FDHT PET) imaging directly targets the androgen receptor and is a promising prognostic and predictive biomarker in metastatic castration-resistant cancer (mCRPC). To optimize [¹⁸F]FDHT PET-CT for diagnostic and response assessment purposes, we assessed how count statistics and reconstruction protocol affect its accuracy, repeatability, and lesion detectability.

Methods: Whole body [¹⁸F]FDHT PET-CT scans were acquired on an analogue PET-CT on two consecutive days in 14 mCRPC patients harbouring a total of 336 FDHT-avid lesions. Images were acquired at 45 min post-injection of 200 MBq [¹⁸F]FDHT at 3 min per bed position. List-mode PET data were split on a count-wise basis, yielding two statistically independent scans with each 50% of counts. Images were reconstructed according to current EANM Research Ltd. (EARL1, 4 mm voxel) and novel EARL2 guidelines (4 mm voxel + PSF). Per lesion, we measured SUVpeak, SUVmax, SUVmean, and contrast-to-noise ratio (CNR). SUV was normalized to dose per bodyweight as well as to the parent plasma input curve integral. Variability was assessed with repeatability coefficients (RCs).

Results: Count reduction increased liver coefficient of variation from 9.0 to 12.5% and from 10.8 to 13.2% for EARL1 and EARL2, respectively. SUVs of EARL2 images were 12.0-21.7% higher than EARL1. SUVs of 100% and 50% count data were highly correlated (R² > 0.98; slope = 0.97-1.01; ICC = 0.99-1.00). Intrascan variability was volume-dependent, and count reduction resulted in higher intrascan variability for EARL2 than EARL1 images. Intrascan RCs were lowest for SUVmean (8.5-10.6%), intermediate for SUVpeak (12.0-16.0%), and highest for SUVmax (17.8-22.2%). Count reduction increased test-retest variance non-significantly (p > 0.05) for all SUV types and normalizations. For SUVpeak at 50% of counts, RCs remained < 30% when small lesions were excluded. Splitting data reduced CNR by median 4.6% (interquartile range 1.2-8.7%) and 4.6% (interquartile range 1.2-8.7%) for EARL1 and EARL2 images, respectively.

Conclusions: Reducing [¹⁸F]FDHT PET acquisition time from 3 min to 1.5 per bed position resulted in a repeatability of SUVpeak (bodyweight) remaining ≤ 30%, which is generally acceptable for response monitoring purposes. However, EARL2 reconstruction was more affected, especially for SUVmax whose repeatability tended to exceed 30%. Lesion detectability was only slightly impaired by reducing acquisition time, which might not be clinically relevant in mCRPC.

Keywords: Count statistics; PET-CT; Reconstruction; [18F]-fluordihydrotestosterone; mCRPC.

Fig. 1

Illustration of a PET image of a typical mCRPC patient with extensive [¹⁸F]FDHT-avid bone metastases reconstructed with a 100% count EARL1, b 50% count EARL1, c 100% count EARL2, and d 50% count EARL2. Axial (left column), coronal (middle column), and sagittal views (right column) are shown

Fig. 2

Schematic representation of assessment of test-retest variability of original 100% count scans and split 50% count scans, respectively. Note that, in contrast with original scans, split scans cannot be directly compared, as this would yield 4 individual comparisons underestimating true test-retest variability

Fig. 3

Liver COV% for 100% count and 50% count EARL1 and EARL2 images

Fig. 4

Correlations between SUVbw of original 100% count scans and split 50% count scans for SUVmean (a, b), SUVpeak (c, d), and SUVmax (e, f). Results from both EARL1 images (a, c, e) and EARL2 images (b, d, f) are shown

Fig. 5

Relative difference (%) between SUVs derived from EARL1 images compared to SUVs derived from EARL2 images as a function of lesion ARTV. a Results from original (100% of counts) scans and b split (50% of counts) images

Fig. 6

Bland-Altman graph of intrascan variability due to 50% count reduction for SUVmean (a, b), SUVpeak (c, d), and SUVmax (e, f). Results from both EARL1 images (a, c, e) and EARL2 images (b, d, f) are shown. Variability was derived from the relative difference in SUV between splits 1 and 2 of each scan on each day. Note that bw or AUC-PP normalization are not reported separately since normalization factors are identical for split 1 and split 2

Fig. 7

Bland-Altman graph of interscan (test-retest) variability of SUVmean (a, b), SUVpeak (c, d), and SUVmax (e, f) normalized to bodyweight at 100% and 50% of counts. Results from both EARL1 images (a, c, e) and EARL2 images (b, d, f) are shown