Baseline Imaging Derived Predictive Factors of Response Following [177Lu]Lu-PSMA-617 Therapy in Salvage Metastatic Castration-Resistant Prostate Cancer: A Lesion- and Patient-Based Analysis

Earlier studies have mostly identified pre-therapeutic clinical and laboratory parameters for the prediction of treatment response to [177Lu]Lu-PSMA-617 in metastatic castration resistant prostate cancer patients (mCRPC). The current study investigated whether imaging-derived factors on baseline [68Ga]Ga-PSMA-11 PET/CT can potentially predict the response after two cycles of [177Lu]Lu-PSMA-617 treatment, in a lesion- and patient-based analysis in men with mCRPC. Included patients had histologically proven mCRPC and a [68Ga]Ga-PSMA-11 PET/CT before and after two cycles of [177Lu]Lu-PSMA-617 treatment. The imaging-based response was evaluated on lesion-level (standardized uptake value (SUV) reduction) and patient-level (total lesion PSMA (TL-PSMA) reduction). In the lesion-level analysis, a clear relationship was found between SUVpeak/max and the imaging-based response to [68Ga]Ga-PSMA-11 PET/CT (most avid lesion SUVpeak/max ≥ 30% reduction) (p < 0.001), with no significant difference in cut-off values between different sites of metastases (i.e., lymph node, bone or visceral metastasis). In patient-level analysis, baseline PSA and SUVpeak values of most avid metastasis were significantly associated with imaging-based response (TL-PSMA ≥ 30% reduction) (p = 0.019 and p = 0.015). In pre-treatment with [68Ga]Ga-PSMA-11 PET/CT, a clear accumulation-response relationship in lesion-level was found for SUVpeak/max in men with mCRPC receiving two cycles of [177Lu]Lu-PSMA-617 treatment. The SUVpeak of the most avid lesion was the only image-derived factor predictive of the imaging-based response at the patient-level.


Introduction
Worldwide, prostate cancer was the third most common diagnosed malignancy in 2020 [1]. The survival rates of prostate cancer are subject to the degree of metastasis. The five-year survival of localized prostate cancer is 100%, however, it falls rapidly to 31% in patients with distant metastases [2].
In the last decade, new treatment options for patients with metastatic, castration resistant prostate cancer (mCRPC) became available, including novel androgen axis drugs (e.g., abiraterone, enzalutamide) and chemotherapy (i.e., docetaxel and cabazitaxel). More recently radioligand therapy with lutetium-177 prostate specific membrane antigen ([ 177 Lu]Lu-PSMA-617) emerged as a promising treatment for advanced prostate cancer [3]. Several studies have demonstrated the safety and efficacy (extended overall survival and improved

Population
Patients referred for and treated with [ 177 Lu]Lu-PSMA-617 were identified retrospectively from a single center from March 2017 to November 2019. Patients were included if they had histologically proven mCRPC and had a [ 68 Ga]Ga-PSMA-11 PET/CT before and after two cycles of [ 177 Lu]Lu-PSMA-617 treatment. The reason why we choose to analyze after two cycles is based on the findings of Ahmadzadehfar et al., who showed that response (PSA decline ≥ 50%) is only or mostly seen after the second cycle of [ 177 Lu]Lu-PSMA-617 treatment [13].
Patients were excluded if the interval between the baseline and post treatment [ 68 Ga]Ga-PSMA-11 PET/CT was more than eight months. Blood testing was performed at the time of admission. The PSMA-617 ligand was obtained from ABX GmbH, Radeberg, Germany. A total of 6.0 or 7.4 GBq [ 177 Lu]Lu-PSMA-617 per 40 to 250 µg peptide was administered intravenously for each cycle, with a planned interval of six weeks.
The need for informed consent was waived by the institutions medical ethics committee for this retrospective study.

Image Acquisition and Reconstruction
Sixty minutes after intravenous administration of 1.5-2.0 MBq/kg [ 68 Ga]Ga-PSMA-11 the imaging was performed from the skull vertex to the mid-thigh (Biograph mCT scanner, Siemens, Erlangen, Germany).

Imaging Analysis
Syngo.via-software (Siemens version 05.01, Erlangen, Germany) was used to establish quantitative image analysis. Based on PERCIST, relevant volumes of interest were (semi)automatically segmented if the standardized uptake value of peak (SUV peak ) was greater than the threshold set by a 3 cm cylindrical volume of interest (VOI) in the aorta with a threshold of 1.5 × aorta peak + 2 × standard deviation [17]. The activity in the blood pool has been shown to be a well-grounded reference region for [ 68 Ga]Ga-PSMA-11 imaging interpretation [18].
Manual adoption was needed if single tumor lesions and organs were not automatically divided based on the set PERCIST criteria.
Segmentations smaller than 0.3 mL where disregarded.
The Syngo.via software only allowed visual validation of a maximum of 50 lesions on the [ 68 Ga]Ga-PSMA-11 PET/CT. It automatically calculated the total amount of [ 68 Ga]Ga-PSMA-11 accumulation of the remaining lesions (>50).
Parameters collected included PSMA tumor volume (PSMA-TV) in mL and TL-PSMA (summation of the entire tumor load within the patient derived from total lesion glycolysis (TLG)). The SUV peak and SUV max of the primary prostate tumor (if in situ), and the SUV peak and SUV max the two most-and least-avid lesions of three different organ categories (lymph nodes, bone and visceral metastasis) were collected. This approach was chosen in order to collect a wide variety of lesion avidity for the lesion-based analysis.
In accordance with the EARL recommendations, the TL-PSMA was calculated by multiplying the SUV peak value with the PSMA-TV (SUV lbm,peak *cm 3 ) per patient. Although the EARL recommendations are used for 18F-FDG (FDG), the used method for the total lesion glycolysis (TLG) will best represent the in vivo distribution of [ 68 Ga]Ga-PSMA-11 by calculating the fractional tumor activity [19][20][21].

Outcomes
The primary outcome of this study was defined as an objective response after two cycles of [ 177 Lu]Lu-PSMA-617 treatment at the lesion-and patient-level. Response evaluation at the lesion-level was based on PERCIST [17]: imaging complete response (iCR); complete resolution of PSMA-tracer accumulation in all lesions, imaging partial response (iPR); more than or equal to 30% reduction of SUV peak , imaging progressive disease (iPD); more than or equal to 30% increase in SUV peak and imaging stable disease (iSD); not qualifying for iCR, iPR, or iPD. The definition of objective response includes iCR + iPR. For the patientbased analysis, the same methodology was used, except the TL-PSMA was used as the distinctive parameter instead of the SUV peak , thus objective response at the patient-level was determined as a reduction of TL-PSMA ≥30% and progressive disease was defined as more than or equal to 30% increase in the TL-PSMA and/or the appearance of new lesions.
Secondary outcomes included a biochemical response after two cycles of [ 177 Lu]Lu-PSMA-617 treatment at the patient-level, defined according to the prostate cancer clinical trial working group 2 and 3 [22,23]. Response definitions: a partial response (bPR) was more than 50% PSA level reduction; progressive disease (bPD) was more than or equal to 25% increase; and a stable disease (bSD) was less or equal to 50% reduction and less than 25% increase of PSA level.
Additionally, clinical, biochemical, imaging, and hematological parameters (Tables 1 and 2,  and Appendix A Table A1) were gathered to investigate potential predictive factors on a patient-level.  Finally, overall survival (OS), defined as after the first cycle of [ 177 Lu]Lu-PSMA-617 treatment to death from any cause, was analyzed on a patient-level.

Statistical Analysis
The software IBM SPSS Statistics version 25.0.0.2 for Windows (IBM, Armonk, NY, USA) and R version 4.0.1 (R Core Team 2020) was used for all analyses (for used R codes see Supplementary Materials). As accumulation measurements (i.e., SUV peak and SUV max ) showed positive skewness, a log-transformation was applied before analyses were executed. Several types of analyses were executed to test different hypotheses. A p-value ≤ 0.05 was considered significant.
For our primary outcome, the imaging-based response on a lesion-level, Receiver Operating Characteristics (ROC) curve analyses were performed for predicting imagingbased response, including the following variables: SUV peak /SUV max all measured lesions together, SUV peak /SUV max lymph node metastases and bone metastases. The Youden's index test and a set minimum specificity of 0.80 were used to determine the optimal cut-off value for binarization of the predictive values.
Mixed-effects models with SUV peak /SUV max as the independent variable, imagingbased response (dichotomized or as categorical variable) as the dependent variable, and a random intercept of SUV peak and SUV max per patient was used to model the effect of imaging-based response on the SUV peak and SUV max values. The random intercept was added to the model to incorporate the anticipated between-patient variation in SUV peak and SUV max levels. The remaining dependent relation between the imaging-based response and SUV peak or SUV max was then modelled as a fixed effect. PSMA-TV in the patient and metastasis type were added to the model as a confounder. In order to test the hypothesis that type of metastasis (i.e., lymph node, bone and visceral metastases) was of influence on the relationship between the individual pre-treatment accumulation and imaging-based response per lesion, metastasis type was added as an interaction to the model, both as a categorical variable and as a dichotomized variable in separate models.
For the imaging-based response on a patient-level, the maximum SUV peak and SUV max values in primary tumor or metastases (lymph node, bone, and visceral metastases) were tested for a relationship with response in logistic regression analysis. This approach was also used for the secondary outcome, biochemical response.
Several variables were tested in logistic regression analysis, univariately, while correcting for tumor load by including baseline PSMA-TV in each model. Each model was tested using likelihood ratio tests, comparing the model including the variable with a model only including baseline PSMA-TV.
Overall survival analysis was done using Cox-proportional hazard models, and Kaplan-Meier survival curves were constructed. The ISUP Gleason score, ECOG performance score, extent of disease, and imaging parameters were included in Cox-proportional hazard regression.

Results
A total of 87 patients were treated with [ 177 Lu]Lu-PSMA-617. Of those patients, 32 were eligible for analysis as illustrated in Figure 1. together, SUVpeak/SUVmax lymph node metastases and bone metastases. The Youden's index test and a set minimum specificity of 0.80 were used to determine the optimal cut-off value for binarization of the predictive values.
Mixed-effects models with SUVpeak/SUVmax as the independent variable, imagingbased response (dichotomized or as categorical variable) as the dependent variable, and a random intercept of SUVpeak and SUVmax per patient was used to model the effect of imaging-based response on the SUVpeak and SUVmax values. The random intercept was added to the model to incorporate the anticipated between-patient variation in SUVpeak and SUVmax levels. The remaining dependent relation between the imaging-based response and SUVpeak or SUVmax was then modelled as a fixed effect. PSMA-TV in the patient and metastasis type were added to the model as a confounder. In order to test the hypothesis that type of metastasis (i.e., lymph node, bone and visceral metastases) was of influence on the relationship between the individual pre-treatment accumulation and imagingbased response per lesion, metastasis type was added as an interaction to the model, both as a categorical variable and as a dichotomized variable in separate models.
For the imaging-based response on a patient-level, the maximum SUVpeak and SUVmax values in primary tumor or metastases (lymph node, bone, and visceral metastases) were tested for a relationship with response in logistic regression analysis. This approach was also used for the secondary outcome, biochemical response.
Several variables were tested in logistic regression analysis, univariately, while correcting for tumor load by including baseline PSMA-TV in each model. Each model was tested using likelihood ratio tests, comparing the model including the variable with a model only including baseline PSMA-TV.
Overall survival analysis was done using Cox-proportional hazard models, and Kaplan-Meier survival curves were constructed. The ISUP Gleason score, ECOG performance score, extent of disease, and imaging parameters were included in Cox-proportional hazard regression.

Results
A total of 87 patients were treated with [ 177 Lu]Lu-PSMA-617. Of those patients, 32 were eligible for analysis as illustrated in Figure 1.  The baseline patients and imaging parameters are summarized in Tables 1 and 2. Baseline hematological parameters and radiopharmaceutical characteristics can be found in Appendix A, Tables A1 and A2. A total of 86 lymph nodes, 119 bone, and 17 visceral metastases were extracted. Table 3 represent the imaging-based response rates on a patientand lesion-level. Figure 2 illustrates the response rates of two patients after two cycle of [ 177 Lu]Lu-PSMA-617 treatment (one responder and one non-responder). Table 3. Imaging-based response on the lesion-and patient-level. Legend: IQR = Inter quartile range, PSMA-TV = PSMA tumor volume, PSMA = Prostate specific membrane antigen, SUV = Standardized uptake value, TL-PSMA = Total lesion PSMA. Table 3. Imaging-based response on the lesion-and patient-level.

Lesion-Level
In ROC analysis (Figure 3), the optimal cut-off value to predict imaging response based on Youden's index was 14.87 for SUV peak (sensitivity = 0.36, specificity = 0.90) and 19.08 for SUV max (sensitivity = 0.40, specificity = 0.89). The cut-off values based on a minimum specificity of 0.80 were 12.07 (for SUV peak ; sensitivity = 0.44) and 15.4 (for SUV max ; sensitivity = 0.49), meaning that of all non-responding tumors, 80% showed accumulation below these cut-off values.

Lesion-Level
In ROC analysis (Figure 3), the optimal cut-off value to predict imaging response based on Youden's index was 14.87 for SUVpeak (sensitivity = 0.36, specificity = 0.90) and 19.08 for SUVmax (sensitivity = 0.40, specificity = 0.89). The cut-off values based on a minimum specificity of 0.80 were 12.07 (for SUVpeak; sensitivity = 0.44) and 15.4 (for SUVmax; sensitivity = 0.49), meaning that of all non-responding tumors, 80% showed accumulation below these cut-off values. The relationship between baseline PET parameters and imaging-based response in linear mixed effects models were significant for both SUVpeak and SUVmax, when testing for dichotomized imaging response (p < 0.001 and p < 0.001) and categorical imaging response (p < 0.001 and p < 0.001; Table 4). On average, in responding tumors a 1.80 (95% CI The relationship between baseline PET parameters and imaging-based response in linear mixed effects models were significant for both SUV peak and SUV max , when testing for dichotomized imaging response (p < 0.001 and p < 0.001) and categorical imaging response (p < 0.001 and p < 0.001; Table 4 (Tables 4 and 5). Coefficients of the fixed effects in mixed-model analysis, with a random intercept (SUV peak /SUV max ) per patient. As the outcome data was log-transformed prior to regression, the exponent of the coefficient can be interpreted as the factor of difference in SUV peak /SUV max between the corresponding response category and the reference category. a p-values for models calculated by the likelihood ratio test between the model and an empty model. Legend: iCR = Imaging complete response, iPD = Imaging progression disease, iPR = Imaging partial response, iSD = Imaging stable disease, SUV = Standardized uptake value. Legend: iCR = Imaging complete response, iPD = Imaging progression disease, iPR = Imaging partial response, iSD = Imaging stable disease, SUV = Standardized uptake value.
The type of lesion (i.e., lymph node, bone, visceral metastasis, or primary prostate) did not alter the found relationships. Figures 4 and 5 shows that lesions with a higher accumulation (SUV peak ) at baseline have a better imaging-based response, with the exception of complete response.  Relationship between metastasis type, response, and accumulation. Numbers in the plot indicate the number of tumors in the corresponding group. Legend: iCR = Imaging complete response, iPD = Imaging progression disease, iPR = Imaging partial response, iSD = Imaging stable disease, SUV = Standardized uptake value.

Patient-Level
Results of the response evaluation on the patient-level are shown in Table 6. Baseline PSA (median 210.0 ng/mL) and SUVpeak most avid metastases had a significant relationship with imaging-based response (OR 2.07, p = 0.019 and OR 1.11, p = 0.015). Imagingbased response was highly associated with biochemical response (p < 0.001). Secondary, no factors were identified having a significant relationship with biochemical response.   Relationship between metastasis type, response, and accumulation. Numbers in the plot indicate the number of tumors in the corresponding group. Legend: iCR = Imaging complete response, iPD = Imaging progression disease, iPR = Imaging partial response, iSD = Imaging stable disease, SUV = Standardized uptake value.

Patient-Level
Results of the response evaluation on the patient-level are shown in Table 6. Baseline PSA (median 210.0 ng/mL) and SUVpeak most avid metastases had a significant relationship with imaging-based response (OR 2.07, p = 0.019 and OR 1.11, p = 0.015). Imagingbased response was highly associated with biochemical response (p < 0.001). Secondary, no factors were identified having a significant relationship with biochemical response. Figure 5. Relationship between metastasis type, response, and accumulation. Numbers in the plot indicate the number of tumors in the corresponding group. Legend: iCR = Imaging complete response, iPD = Imaging progression disease, iPR = Imaging partial response, iSD = Imaging stable disease, SUV = Standardized uptake value.

Patient-Level
Results of the response evaluation on the patient-level are shown in Table 6. Baseline PSA (median 210.0 ng/mL) and SUV peak most avid metastases had a significant relationship with imaging-based response (OR 2.07, p = 0.019 and OR 1.11, p = 0.015). Imaging-based response was highly associated with biochemical response (p < 0.001). Secondary, no factors were identified having a significant relationship with biochemical response.
During follow-up, 28 (87.5%) patients were found to have died with a median overall survival of ten months ( Table 2). Log-rank testing showed that patients with an ECOG performance score of zero or one have a significant better survival rate than patients with an ECOG performance score of two (p = 0.033), the same for patients with >50% PSA reduction in comparison to no PSA reduction of >50% (p = 0.05) and for patients with ≥30% TL-PSMA reduction in comparison to no reduction of ≥30% (p = 0.048). Patients who had the primary tumor in situ had a better overall survival (p = 0.006) ( Figure 6). The factors ISUP, presence of visceral metastases, bone metastases or lymph node metastases, baseline TL-PSMA, baseline PSMA-TV and most avid lesion SUV peak , SUV max did not shown any significance.   In the univariate Cox-regression analyses, patients with an ECOG performance score of two and an unknown Gleason (ISUP) score had a significant hazard ratio (HR) ( Table 7), but overall, these variables were not significantly associated with survival (p = 0.104 and p = 0.386). Biochemical response was significantly associated with survival (HR 0.43, p = 0.047; Table 7).  In the univariate Cox-regression analyses, patients with an ECOG performance score of two and an unknown Gleason (ISUP) score had a significant hazard ratio (HR) ( Table 7), but overall, these variables were not significantly associated with survival (p = 0.104 and p = 0.386). Biochemical response was significantly associated with survival (HR 0.43, p = 0.047; Table 7).

Discussion
This study evaluated the potential of imaging-derived factors on [ 68 Ga]Ga-PSMA-11 PET/CT to predict the response in a lesion-and a patient-based analysis, in men with mCRPC receiving two cycles of [ 177 Lu]Lu-PSMA-617 treatment. In the lesion-level analysis, a clear relationship was found between pre-therapeutic accumulation (SUV peak and SUV max ) and imaging-based response on [ 68 Ga]Ga-PSMA-11 PET/CT with no preference or difference for either, primary tumor, lymph node, bone or visceral metastasis.
Interestingly, in the lesion-level analysis, a contradictory lower SUV peak at baseline was seen in lesions with iCR. An explanation might be the threshold method based on PERCIST and lesion selection criteria in this study. If, at follow-up imaging, the lesion SUV peak was under the threshold, it was set to zero (being iCR), even when visually some accumulation might still be present. An additional explanation can be found in the set threshold and making the accumulation of the least avid lesion depended on the blood pool activity. However, to address a large number of lesions with a wide variety of intensities, this approach was deliberately chosen to gain more insight in lesion-based response. A third explanation is the influence of partial volume effect on small lesions, potentially overestimating objective response.
In the clinical setting, a difference was noticed in the objective response between different lesion types (e.g., prostate, lymph node, bone or visceral), however, the results of this study show no difference between lesion type (Figures 4 and 5). Thus, making a distinction between lesion types seems irrelevant for patient selection prior to 177 Lu-PSMA-617. To the best knowledge of the authors, this is the first study evaluating the response on individual lesion-level with imaging-derived predictive factors on [ 68 Ga]Ga-PSMA-11 PET/CT, thus no comparison with existing literature can be made.
On the patient-level, the pre-therapeutic imaging-derived predictive factor, SUV peak of the most avid metastases was significantly associated with the imaging-based response (TL-PSMA ≥ 30% reduction). No other study has evaluated the imaging-based response based on PERCIST. Hofman et al. [5] and Sartor et al. [3], however, used the RECIST criteria, but in comparison to PERCIST (which looks at accumulation reduction); RECIST only uses single dimension size changes and has severe limitations in measuring bone or bone marrow disease [17]. On the other hand, evaluating tumor response with [ 68 Ga]Ga-PSMA-11 PET/CT (via the same mechanism of the treatment itself) can also be debated, as potential non-PSMA avid disease will not be evaluated.
There are studies evaluating imaging-derived predictive factors with biochemical response as outcome, although the results are contradictory. Some did not find any significant imaging-derived predictive factor (e.g., SUV max and SUV mean , total tumor load, number of metastatic lesions, and sites of disease) for biochemical response [11,24], while some did (e.g., SUV max < 45 of the most avid lesion, SUV mean ) [10,25]. An explanation can be found in the difference in the number of cycles used, population size, the used therapeutic radiopharmaceuticals, and the amount of activity, thereby making their results difficult to compare and to interpret.
In the survival analyses, patients with an ECOG performance score of zero and one had a significant better OS than patients with an ECOG performance score of two, in line with previous findings [7]. Furthermore, patients with a biochemical response (>50% PSA reduction) had a better OS, compared to biochemical non-responders. This is in contrast to the findings of Ahmadzadehfar et al. [7] and Rahbar et al. [9], who did not find a significant difference in 100 and 104 mCRPC patients treated with [ 177 Lu]Lu-PSMA-617 between biochemical responders and non-responders concerning OS. The difference can possibly be explained by differences in sample size, population heterogeneity and selection bias by the range in the number of given [ 177 Lu]Lu-PSMA-617 cycles in both studies: namely two to six in this study versus one to eight cycles.
This study has several limitations: first, the retrospective design and resulting missing data. Second, the small sample size ( Figure 1) and population heterogeneity limits the ability to draw definite conclusions on patient-based analyses. But in the lesion-based analysis, the sample size was a total of 237 lesions, however, only the two most avid and two least avid lesions were selected to address many lesions with a wide variety of intensities, introducing a selection bias. Third, no volumes of the individual lesions were measured, introducing bias by partial volume effects. This could have influenced the response rate on the lesion-level analysis, as a lower PSMA-TV with the same SUV peak is more prone for iCR than a higher PSMA-TV with the same SUV peak . On the patient-level, however, baseline TL-PSMA and PSMA-TV had no significant influence on the imagingbased response rate. Fourth, PERCIST for the lesion-based response evaluation is not validated for PSMA PET/CT. However, it is already broadly available in clinical practice and easy to apply [15,17].
Other studies used a maximum intensity threshold with SUV max for tumor segmentation [20,26]. In this study, we chose to use SUV peak adapted from PERCIST for tumor segmentation as this limits the influence of noise on quantification [27].
In current practice, patients are only eligible for [ 177 Lu]Lu-PSMA-617 treatment if sufficient tracer accumulation is observed on PSMA PET/CT. However, the definition of sufficient tracer accumulation is still a topic of discussion. Currently, it is based on literature on peptide receptor radionuclide therapy (PRRT), as was also used in the VISION trial [28][29][30]: accumulation in tumor sites must at least be higher than physiological accumulation in normal liver tissue, to ensure a certain efficacy. The included patients in this study all met this specific criterion. Still, there were some non-responders (iPD and bPD) in this study (5/32; 16% and 5/30; 17%), in line with the findings in the VISION trial [3], thereby indicating that the decision whether or not an individual patient is eligible for [ 177 Lu]Lu-PSMA-617 based on the visual assessment of accumulation alone compared to healthy liver tissue accumulation remains questionable. The results in this study indicate that tracer accumulation based on SUV peak (>14.87) or SUV max (>19.08) in a lesion can be helpful to determine if a certain lesion will or will not respond, based on a broadly available, internationally accredited image reconstruction method (EARL) [14]. In case, when all or the majority of metastases within a patient are below these thresholds, an alternative treatment may be more beneficial, subsequently, improving patient selection for 177 Lu-PSMA-617 based on available pre-treatment [ 68 Ga]Ga-PSMA-11 imaging.
The results of this study illustrate the potential of response prediction by pre-treatment [ 68 Ga]Ga-PSMA-11 PET/CT quantification, using widely available image reconstruction parameters (EARL) and software packages enabling (semi-automated) PERCIST assessments. The findings of this study need to be validated in larger cohorts and future prospective studies.

Conclusions
On pre-treatment [ 68 Ga]Ga-PSMA-11 PET/CT, a clear accumulation-response relationship in lesion-level analyses has been found for SUV peak and SUV max in men with mCRPC receiving two cycles of [ 177 Lu]Lu-PSMA-617 treatment. On a patient-level analysis, SUV peak of the most avid lesion was the only image-derived factor predictive of imaging-based response.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/biomedicines10071575/s1, PSMA-PET response-mixed model analysis.  Institutional Review Board Statement: Ethical review and approval were waived by the institutional medical ethics committee for this study, due to the retrospective nature of the study.

Informed Consent Statement:
The need for informed consent was waived by the institutional medical ethics committee.

Data Availability Statement:
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Conflicts of Interest: MGEHL has acted as consultant for BTG/Boston Scientific and Terumo/Quirem
Medical and receives research support by Novatis/AAA. AJATB has acted as consultant for BTG/Boston Scientific and Terumo/Quirem Medical. All other authors declare that they have no conflict of interest.  Legend: Ga = Gallium, Lu = Lutetium, IQR = Inter quartile range, PSMA = Prostate specific membrane antigen.