Patient population
This study included consecutive patients diagnosed with metastatic and progressive or recurrent GEP-NET, treated with at least one fraction of [177Lu]Lu-DOTA-TATE PRRT. Eligibility criteria for the treatment were radiologic, symptomatic or biochemical progression despite optimal conventional treatment, and a high expression of SSTR (grade 3–4) as assessed on [68Ga]Ga-DOTA-TOC PET/CT scans. Histological sampling was performed prior to the treatment to establish tumour grade and Ki-67 index. Adequate hepatic, renal and hematologic function are also critical prerequisites for treatment. This study was approved by the local data protection officer at Oslo University Hospital, and informed consent was waived. Data was handled according to relevant regulations.
Image Acquisition
Pre-therapeutic PET/CT imaging (example in Fig. 1A and 1C) was performed on a GE Discovery MI, 78 minutes (range: 54–96 minutes) post administration a nominal activity dosage of 150 MBq [68Ga]Ga-DOTA-TOC. The acquisition time was 100 s per bed with 50% overlap. The images were gated using the Varian trigger system and reconstructed in the expiratory phase (Q.Static). The images were reconstructed with a Bayesian penalised likelihood reconstruction (Q.Clear) with a beta factor of 500, on a 384x384 matrix of 1.82 mm pixels with a slice thickness of 2.79 mm. The CT scans were acquired at 120 kV and a dose modulation noise index of 34.5, and reconstructed with an iterative reconstruction (ASIR-V) with 40% filtering on 512x512 matrices. The pixel size was 0.84 mm with slice thickness 0.625 mm.
[177Lu]Lu-DOTA-TATE SPECT/CT imaging (example in Fig. 1B, 1D and 1E) was on average performed 23.1 hours (range 21.0-26.8) and 167.0 hours (range 119.1-218.8 hours) post administration (t24 and t168, respectively) of the first treatment cycle for all patients. An additional image (t4) on average 4.7 hours (range 3.4–6.5) was acquired for nine of the patients. The average administered activity was 7574 MBq (range 7247–7907 excl. one patient given 4195 MBq). The imaging was performed on a GE Discovery 670 scanner with a medium energy, general purpose collimator. The images were acquired with an energy window set at 208 keV +/- 10% with 120 projections of 30 seconds frame duration, and two adjacent scatter windows of +/- 5%. A scatter- and attenuation-corrected reconstruction with 4 iterations and 8 subsets without resolution recovery was performed with the vendor software (Xeleris, GE Healthcare). Matrix size was128x128 and isotropic voxels of 4.42 mm were reconstructed. The CT scans were acquired at 120 kV and a dose modulation noise index of 26, and reconstructed with an iterative reconstruction (ASIR) with 40% filtering on 512x512 matrices. The pixel size was 0.98 mm with slice thickness 2.5 mm.
[68Ga]Ga-DOTA-TOC PET Measurements
Tumours which could be visually matched between all images (PET and SPECT at t24 and t168), with negligible overlap with adjacent activity sources, were included for analysis. A manual identification of tumours confirmed by a nuclear medicine specialist was followed by a fixed threshold method for measurements, using a threshold of 40%24 of the SUVmax. Both SUVmax, SUVmean, and the volume defined by the threshold; referred to as the somatostatin receptor expressing tumour volume (SRETV) from here on, were recorded.
[177Lu]Lu-DOTA-TATE SPECT Measurements and Tumour Dosimetry
The tumours were delineated using 40% threshold determined by volume matching in a phantom study, from which recovery coefficients were also found (supplementary Fig. 1). The VOI-derived counts were converted to activity by a system specific calibration coefficient of 8.4 cps/MBq. Time-activity curves (TACs) were calculated by assuming a linear activity build-up from time of administration to the first measuring point (t24), followed by a mono-exponential curve fit based on t24 and t168. The TAC was integrated to yield the time-integrated activity (Ã). Tumour mass, m, was estimated by assuming a tissue mass density of 1 g/ml. The absorbed dose was calculated by assuming local deposition and a dose factor DF = 0.083 Gy∙g/MBq∙h, including all non-penetrative radiation from 177Lu.
$$AD= \frac{\tilde{A}}{m}DF$$
1
Prediction Analysis
A linear relation was assumed between [68Ga]Ga-DOTA-TOC SUVmean and [177Lu]Lu-DOTA-TATE activity concentration per administered activity at t24 (\({A}_{conc,24}/{A}_{adm}\)), as a measure of uptake correlation.
$${A}_{conc,24}/{A}_{adm}=SU{V}_{avg}\cdot \alpha +\beta$$
3
The absorbed dose prediction model (Eq. 4) was based on SUV measurements and administered activity.
$$A{D}_{pred}\left({SUV}_{avg}\right) = ({SUV}_{avg}\cdot \alpha +\beta )\cdot {A}_{adm}\cdot DF\cdot \tau$$
4
The fitting parameters α and β were derived from the linear relationship between \({A}_{conc,24}/{A}_{adm}\) and SUVmean. \(\tau\) is a constant that represents the shape of the TAC based on the patient population average of t24.pop and teff.pop:
$$\tau =\frac{{t}_{24,pop}}{2}+\frac{{t}_{eff,pop}}{\text{ln}\left(2\right)}$$
5
Group analysis
An analysis to investigate differences in absorbed dose and differences between predicted and actual absorbed dose for multiple tumour groups was conducted. Tumours were grouped by anatomical site (hepatic, lymphatic and skeletal lesions were compared), SUV-parameters and tumour mass. Two approaches were explored; one where tumours were divided into groups of equal size, and one where equal range of the parameter were grouped.
Statistics
A Shapiro-Wilk normality test and visual inspection of the probability plots of the residues were performed to test for normality. Pearson correlation was performed to test linear correlation between parameters. The linear correlation was evaluated in terms of the coefficient of determination, R2. A significance level of 0.05 was used. Relative error (\({\varDelta }_{AD}\)) was used to determine the accuracy of the prediction model, defined by the relative difference between predicted, ADpred, and actual absorbed dose:
$${\varDelta }_{AD}=100\%\bullet ({AD}_{pred}-AD)/AD$$
6
The accuracy of the prediction model was further explored by testing the ability to predict if the tumour absorbed dose was higher or lower than the population median using receiver operating characteristics analysis.
The group differences of relative error between predicted and measured absorbed dose were tested with a one-way analysis of variance (ANOVA)-test and significance of individual group differences were conducted using the Tukey-test. The difference of absorbed dose between groups was also tested.
All statistical analyses were conducted with the SciPy Statistical functions module in Python 3.8 (Python Software Foundation). The probability plot of the residuals was created with the SciPy stats probplot function against the normal distribution.