Potential of asphericity as a novel diagnostic parameter in the evaluation of patients with ⁶⁸Ga-PSMA-HBED-CC PET-positive prostate cancer lesions

This retrospective study was approved by the local ethics review board. The local database was screened for patients, who received a ⁶⁸Ga-PSMA-PET combined with computed tomography (CT) and a 3 T MRI of the prostate within 110 days for staging of suspected primary PC. MRI was used as reference standard for the definition of the primary tumor and to evaluate our delineation process. The proximity of the ⁶⁸Ga-PSMA PET/CT to the MRI was a requirement to ensure that the lesion which was evaluated in ⁶⁸Ga-PSMA PET reflects the primary PC lesion. We extracted 691 patients from our imaging database, who underwent ⁶⁸Ga-PSMA-PET/CT in-between Oct. 01, 2013 and Feb. 01, 2017. MRI data of the prostate was available in 169 cases, excluding 522 cases without a MRI of the prostate. In our institution, 3 T MRI has established itself as the routine MR examination for the evaluation of patients with prostate cancer. Twenty-six cases were excluded as only 1.5 T MRI datasets were available. Ninety patients were excluded as the delay between ⁶⁸Ga-PSMA-PET/CT and 3 T MRI was more than 110 days. All PSMA-positive lesions had to be confirmed through core needle biopsy, which was not available for this study in 16 patients. The final cohort of 37 patients had a mean age of 71.3 ± 7.5 years and received both scans within 50.2 ± 32.5 days. Core needle biopsy and ⁶⁸Ga-PSMA-PET/CT were performed within 74.2 ± 80.2 days. A detailed overview regarding the patient characteristics are summarized in Table 1.

Elution of ⁶⁸Ga was performed using a standard ⁶⁸Ge/⁶⁸Ga generator (Eckert and Ziegler) [22]. Next, PSMA-HBED-CC (ABX GmbH, Radeberg, Germany) was labeled with ⁶⁸Ga [23, 24].

PET/CT imaging was performed 75.4 ± 27.5 min after intravenous injection of 122.4 ± 19.7 MBq of [⁶⁸Ga]-PSMA-HBED-CC. A 3D acquisition mode was used on a Gemini TF 16 Astonish PET/CT scanner (Philips Medical Systems) [25]. Default parameter settings were used in the system software to reconstruct the transaxial slices (144 × 144 voxels, 4 mm³). Immediately before the PET scan, a low-dose CT was acquired for attenuation correction (120 kVp, 30 mAs) and anatomical mapping.

Definition of ASP:ASP is defined as a marker for non-spherical tumor volumes. A sphere has the smallest possible surface S for a given volume V for which ASP = 0 by definition [15]. For non-spherical lesions, ASP > 0 provides a quantitative measure for deviation of spherical shape. For example, an ASP of 30% means a 30% larger surface than a sphere with the same volume [19]. A detailed description of the definition was published earlier [15].

The program Visage 7.1 (Visage Imaging) was used prior to the delineation process for evaluation of the ⁶⁸Ga-PSMA-PET/CT and the MRI data. In case of multiple foci or lesions with unclear borders in PET, a simultaneous evaluation of the MRI data in Visage 7.1 supported the detection/delineation of the primary PC lesion. Following the detection of the primary lesion within the prostate using the rover software (ABX GmbH, Radeberg, Germany), a 3D mask was placed around the volume of interest (VOI). Rover uses a specific algorithm that delineates the tumor automatically. This is achieved by adaptive thresholding and taking the background signal of the surrounding tissue into account [20]. In some cases, the strong radiotracer signal from the bladder was interfering with a fully automatic delineation of the tumor. Therefore, the tumor delineation was inspected in axial, coronal, and sagittal planes in all cases and the VOIs were corrected manually, if necessary. Computed parameters of the VOIs included the ASP, SUVmax, SUVmean, SUVpeak and the viable tumor volume (VTV). Figures 1 and 2 show examples of the delineation of the tumor using the rover software in patients with different Gleason scores (GSc).

Accumulation of ⁶⁸Ga-PSMA-HBED-CC in PC cells depends on the PSMA expression on the cell surface. PSMA is highly overexpressed in PC cells, resulting in a strong signal in PET [26]. The accumulation of ⁶⁸Ga-PSMA-HBED-CC is reduced in non-diseased prostate tissue, expressing lower PSMA levels on their cell surface. This leads to a delineation of a VTV, at the location at which ⁶⁸Ga-PSMA-HBED-CC internalization is highly active. Derived from the “total lesion glycolysis” used in ¹⁸F-FDG-PET/CT, a further parameter used in this study was the “total lesion binding” rate (TLB). The TLB was defined as the product of the SUVmean and the VTV of a PC lesion.

All 37 patients underwent core needle biopsy of the prostate for a histopathological characterization of tumor malignancy. The GSc takes into account that malignancy signs such as cell size, nucleus size, nucleus to cytosol ratio, abnormal mitosis, and necrosis affect the grading. The final score is an addition of the most common found tumor grade and the highest found tumor grade for all tissue samples [27].

TNM is a clinical used score for staging, for determination of prognosis and treatment planning. T-Stage describes the tumor size and tumor infiltration in surrounding tissue. N-stage indicates the presence of lymph node metastases; M-stage describes the presence of remote metastases outside the prostate [28].

The risk stratification for progression of PC can be measured with the D’Amico classification tool. It takes PSA blood level, clinical tumor size (through endorectal examination), and GSc (core needle biopsy) into account. It classifies patients according to low, intermediate, or high risk for early metastasis and increased tumor aggressiveness [29].

Prostate imaging reporting and data system (PI-RADS) is a score used for standardized reporting of clinical findings in magnetic resonance imaging examinations of the prostate. In 2014, an update towards version 2 was released. MRI sequences included in PI-RADS v2 evaluation, which was used for all patients in this study, are high-resolution T2-weighted imaging, diffusion-weighted imaging, and dynamic contrast-enhanced imaging sequences. For each sequence, a score of 0–5 indicates the probability of a clinical significant PC lesion leading from improbable to highly suspicious in score 5. Every lesion is scored in the three sequences resulting in three sub-scores. The resulting final PI-RADS score is a summarized score [30, 31].

The SUV is a degree of tracer uptake in a specific region of interest (ROI) or VOI. It is calculated as the product of the activity concentration (Bq/g) and the patient’s weight (g) divided through the applied dose (Bq). SUVmax, SUVmean, and SUVpeak can be calculated for every ROI or VOI [32]. In addition to maximum and mean SUV, SUVpeak was computed as the mean value of a 3D sphere with a diameter of approximately 1.2 cm centered at the VOI maximum. All parameters were computed using the rover software.

Descriptive statistics, correlations, and scatter plots were computed using MedCalc Statistical Software version 17.2 (MedCalc Software bvba, Ostend, Belgium; http://​www.​medcalc.​org; 2017). All univariate correlations including ordinal variables were tested using Spearman’s rank correlation method. Pearson’s correlation method was used for metric variables. A p value < 0.05 was considered statistically significant. The polytomous universal model implemented in the statistic software IBM SPSS (version 24) was used for the ordinal logistic regression. Ordinal regression models the propensity of the first ranked state against all higher ranked states. This is repeated for the second and ongoing ranked states resulting in k–1 intercept parameters for k ordinal levels. We included the independent variables ASP, VTV, and TLB in the model. Ordinal regression is preferable, when the outcome consists of several discrete but ordered states instead of the assumption of a continuous dependent variable in linear regression. The group-based analysis for GSc was tested using Bonferroni corrected t-tests in SPSS. ASP probability was visualized using R software (Version 3.2.5, Vienna, Austria, http://​www.​R-project.​org). Variables are reported as mean as well as standard deviation (SD).

Patients in this study demonstrated GSc ranging from 6 to 10 with a median of 8. Computing of ASP resulted in an overall mean 23.2 ± 10.1% ranging from 5 to 46.5% (CI 19.8–26.5). A close correlation was found between ASP and GSc (rho = 0.88; p < 0.05; CI 0.78–0.94). Twelve patients with GSc of 6–7 demonstrated an average ASP of 11.9 ± 4.8% (range 5.0–18.6%). Fifteen patients with GSc of 8 demonstrated an average ASP of 25.5 ± 4.8% (range 18.9–33.9%). Ten patients with GSc of 9–10 demonstrated an average ASP of 33.3 ± 6.8% (range 20.8–46.5%). Group-based analysis showed significant (p < 0.05) differences in ASP levels for Gleason 6–7 vs. Gleason 8 and for Gleason 8 vs. Gleason 9–10. The bar chart in Fig. 3 demonstrates the subgroup analysis. The scatter plot shows the regression with the associated 95% confidence interval. The correlation was shown to be significant (R ² = 0.84, p < 0.05).

Ordinal logistic regression computed statistically significant (p < 0.05) effects to predict GSc based on given ASP. Threshold ASP values were computed, for the differentiation of the different GSc. Threshold values for a change from subgroup Gleason 6–7 to Gleason 8 was computed at an ASP of 5.6% (CI 7.3–29.2%) and probable change from subgroup Gleason 8 to Gleason 9–10 at an ASP of 30.4% (CI 13.2–48.9%). Grouped ranges were 0–5.6% for Gleason 6–7, 5.6–30.4% for Gleason 8 and ≥ 30.4% for Gleason 9–10. These results are summarized in Fig. 4.

ASP and D’Amico classification showed a moderate correlation (rho = 0.6; p < 0.05; CI 0.32–0.78). Low risk presented an ASP of 5.0%, intermediate risk demonstrated an average ASP of 15.2 ± 2.1% (range 11.7–17.1%), and high risk presented an average ASP of 27.2 ± 8.5% (range 5.2–46.5%). No significant (p > 0.05) correlation was measured between ASP and N-stage (rho = − 0.07; p > 0.05; CI − 0.51–0.4). Additionally, a non-significant (p > 0.05) weak correlation was seen between ASP and the PI-RADS score (rho = 0.26; p > 0.05; CI − 0.11–0.57). All results are summarized in Table 2.

The study cohort presented an average VTV of 12.3 ± 11.3 cm³ ranging from 0.8 to 54.1 cm³ and a moderate correlation could be demonstrated for VTV and GSc (rho = 0.51; p < 0.05; CI 0.23–0.72). VTV averages for GSc 6 were 6.1 ± 7.6 cm³ (range 0.8–17.3 cm³); for GSc 7, 7.5 ± 6.8 cm³ (range 2.3–23.6 cm³); for GSc 8, 11.4 ± 12.5 cm³ (range 2.9–54.1 cm³); for GSc 9, 23.0 ± 9.0 cm³ (range 10.8–39.1 cm³); and for GSc 10, 7.8 ± 6.7 cm³ (range 3–12.5 cm³). VTV presented a moderate association to the respective D’Amico classification (rho = 0.49; p < 0.05; CI 0.17–0.71). VTV was 3.9 cm³ for low risk; intermediate risk scored patients showed an average VTV of 4.7 ± 2.5 cm³ (range 2.3–8.9 cm³). Patients scored with high risk of progression demonstrated an average VTV of 14.6 ± 11.9 cm³ (range 2.9–54.1 cm). No significant (p > 0.05) correlation was found for VTV to N-stage (rho = − 0.06; p > 0.05; CI − 0.5–0.4). VTV presented a non-significant (p > 0.05) weak correlation to the PI-RADS score (rho = 0.33; p > 0.05; CI − 0.04–0.62). All results are summarized in Table 2.

TLB presented an average of 99.2 ± 136.3 ranging between 3 and 584.3 showing statistical significant (p < 0.05) weak correlations to GSc (rho = 0.43; p < 0.05; CI 0.12–0.66) and to D’Amico classification (rho = 0.38; p < 0.05; CI 0.04–0.64). No correlation was found towards N-stage (rho = 0; p > 0.05; CI − 0.45–0.45) and a non-significant (p < 0.05) weak correlation was seen for TLB to the PI-RADS score (rho = 0.3; p > 0.05; CI − 0.07–0.6). All results are summarized in Table 2.

SUVmax of all PC lesions presented an average of 18.1 ± 19.4 ranging between 3.3 and 83.2 and demonstrated non-significant (p > 0.05) weak correlations to GSc (rho = 0.29; p > 0.05; CI − 0.04–0.56) and D’Amico classification (rho = 0.29; p > 0.05; CI − 0.06–0.58). A non-significant (p > 0.05) very weak negative association was seen for SUVmax to N-stage (rho = − 0.14; p > 0.05; CI − 0.56–0.33) and a non-significant (p > 0.05) weak correlation towards PI-RADS score (rho = 0.22; p > 0.05; CI − 0.15–0.54). With an average of 6.1 ± 3.2 ranging between 2.2 and 16.8, SUVmean presented comparable non-significant (p > 0.05) weak to very weak correlations to GSc (rho = 0.24; p > 0.05; CI − 0.09–0.53), D’Amico classification (rho = 0.2; p > 0.05; CI − 0.15–0.51), N-stage (rho = 0.13; p > 0.05; CI − 0.35–0.55), and PI-RADS score (rho = 0.1; p > 0.05; CI − 0.27–0.44). SUVpeak presented an average of 15.3 ± 15.9 ranging between 3.1 and 76.4 with non-significant (p > 0.05) weak correlations to GSc (rho = 0.3; p > 0.05; CI − 0.03–0.57) and D’Amico classification (rho = 0.3; p > 0.05; CI − 0.05–0.58). No correlation was seen towards N-stage (rho = − 0.05; p > 0.05; CI − 0.49–0.42) and a non-significant (p > 0.05) very weak correlation to PI-RADS score (rho = 0.13; p > 0.05; CI − 0.24–0.47). All results are summarized in Table 2.

PSA blood samples were taken in mean 16.2 ± 25.6 days ranging 0 to 84 days and demonstrated an average of 17.7 ± 21.5 ng/ml ranging 0.23 to 116 ng/ml. PSA showed non-significant (p > 0.05) weak to very weak correlations to GSc (rho = 0.27; p > 0.05; CI − 0.06–0.55) and to D’Amico classification (rho = 0.15; p > 0.05; CI − 0.21–0.47). Additionally, no correlation was seen for PSA to N-stage (rho = 0.07; p > 0.05; CI − 0.4–0.51) and a weak non-significant (p > 0.05) correlation to PI-RADS score (rho = 0.12; p > 0.05; CI − 0.25–0.46). All results are summarized in Table 2.

Aggressiveness of tumor behavior, therapy response and overall patient survival is known to be associated with the heterogeneity of the tumor [33, 34]. The parameter ASP enables the quantification of the associated spatial irregularities in PET datasets. Previous studies already investigated the prognostic value of the ASP in certain types of head and neck cancers and non-small-cell lung cancer (NSCLC) [15, 19, 21, 35]. The ASP was found to be an independent predictor of outcome in head and neck cancer patients undergoing pre-therapeutic ¹⁸F-FDG-PET/CT. ASP measurements of the ¹⁸F-FDG uptake also improved the prediction of tumor progression. These previous studies reported a moderate correlation between ASP and MTV, while no correlation between ASP and SUVmax were measured [19, 35]. Additionally, a recent study demonstrated a significant association between progression-free survival and overall survival based on the assessment of the ASP [15].

Comparable results were published for NSCLC, in which the ASP provided a higher prognostic value for progression-free survival and overall survival in NSCLC patients compared to SUVmax, MTV, and another parameter of spatial heterogeneity called solidity [19]. Moderate associations were found between ASP and MTV; no correlation was measured between SUVmax and ASP [19]. Additionally, correlations of ASP with histopathology and with the expression of the tumor proliferation markers KI-67 and epidermal growth factor receptor (EGFR) in NSCLC were found [35].

These previous studies introduced the ASP as a promising novel parameter for the non-invasive characterization of tumors. Additionally, the ASP could represent a strong predictive parameter regarding the overall survival in these patient collectives [15, 19].

Various radiotracers have been tested for the local staging of PC. One of these tracers is ¹⁸F-FDG. Its accumulation is based on an increased glucose metabolism of cancer cells due to an overexpression of hexokinase. ¹⁸F-FDG uptake was shown to be increased in benign prostate tissue, including prostate hyperplasia, as well as in PC cells [11]. Sensitivities of up to 64% for detection of primary PC were reported [36]. The limited performance of ¹⁸F-FDG for the primary PC diagnosis could be associated with the relatively low metabolic rate of PC and the lack of patient selection in these previous studies [6, 8‐10, 37].

A different tracer that demonstrated potential for the detection of PC is ¹⁸F-C. The use of choline-based tracers is dependent on phosphorylcholine turnover in PC cells. Most studies, however, reported limited sensitivities, especially for the primary diagnosis of PC [6, 7, 11]. A further tracer that was evaluated in this context is ¹¹C-acetate. Its uptake is a result of an increased lipid synthesis in tumor cells [38]. Even though its uptake is not limited to PC cells, this radiotracer was shown to be superior to ¹⁸F-FDG for the detection of PC lesions [39]. ¹¹C-MET and ¹⁸F-FDHT have also been evaluated for the staging of PC. ¹¹C-MET targets the increased amino-acid transport of methionine for protein synthesis in cancer cells. ¹⁸F-FDHT targeting is based on an overexpression of the androgen receptor. Limitations of these tracers include the lack of studies regarding their diagnostic value [6].

These previous studies demonstrated that novel more specific tracers and in vivo parameters are needed for an improved in vivo detection and characterization of PC.

PSMA is significantly over-expressed in PC cells and over-expression increases with more advanced tumor stages [26]. Binding of ⁶⁸Ga-PSMA-HBED-CC leads to receptor internalization and tracer accumulation. It is important to mention that PSMA avid tissue can be found throughout the body since it is a zinc-dependent exopeptidase with glutamate carboxypeptidase activity [24]. A recent study demonstrated promising sensitivity, specificity, and accuracy rates of 65.9, 98.9, and 88.5% for the detection of high-risk PC using ⁶⁸Ga-PSMA-HBED-CC. However, a reliable detection can be challenging, as up to 8.4% of all primary tumors showed no tracer accumulation [13, 40]. Another recent study investigated the intensity of tracer accumulation in 90 patients using ⁶⁸Ga-PSMA-HBED-CC. It was demonstrated that the SUVmax of primary PC was significantly higher in GSc > 7 compared to GSc < 7. Additionally, it was shown that a PSA value > 10 ng/ml was an associated with a significantly higher tracer uptake compared to a PSA value < 10 ng/ml [41]. Other studies focused on the diagnostic accuracy of recurrent PC using ⁶⁸Ga-PSMA-HBED-CC. These studies reported sensitivities and specificities up to 80 and 97% for the detection of recurrent PC [42‐45]. PSA blood levels in biochemical recurrence correlated with positive findings, even in patients with low PSA levels (< 1 ng/ml) [14, 43, 46].

To the best of our knowledge this was the first study which combined ⁶⁸Ga-PSMA-PET with the evaluation of the ASP in patients with ⁶⁸Ga-PSMA-HBED-CC-positive PC lesions. Previous studies have, as described earlier, focused on the investigation of the ASP in combination with other radiotracers, such as ¹⁸F-FDG-PET.

The current study demonstrated that the ASP derived from ⁶⁸Ga-PSMA-PET enables a distinction between patient groups with different GSc. Additionally, a correlation of the ASP with the GSc, based on core needle biopsy, was found. The findings in our studies are in line with previous studies, in which a correlation between the ASP and the histopathological staging of NSCLC was demonstrated [35]. In the current study, no significant correlation between SUVmax and ASP was measured [15, 19, 35]. Furthermore, no significant correlation was found between SUVmax and GSc. In contrast to previous studies, this study did not demonstrate a correlation of ASP with the PET tumor volume. This could be explained by smaller VTV of PC lesions in comparison to lesions of NSCLC and head and neck cancer. Our study did not show statistically significant associations of the ASP to the PI-RADS score.

Further prospective studies and a higher number of patients are now warranted to investigate the potential of the ASP in the staging process of PC patients.

This study is limited by its retrospective study design. Only a relatively small patient cohort with clustered GSc was investigated. In case of multiple lesions, PC lesion selection on PET was based on the evaluation of the MRI data. If the strong radiotracer signal from the bladder was interfering with an automatic delineation of the tumor, the tumor delineation was inspected in axial, coronal, and sagittal planes and VOIs were corrected manually, if necessary. Although several viable automated algorithms have been described, the VTV is presently still determined by manual delineation in a high number of institutions [26, 47‐54]. Manual delineation is prone to intra- and interobserver variability as well as to potentially size- and background-dependent bias if fixed absolute or relative thresholds are used.