PI-RADS-Based Segmented Threshold of PSMA-PET SUVmax Is Better than Traditional Fixed Threshold for Diagnosing Clinically Significant Prostate Cancer Especially for PI-RADS 3 Lesions

Our purpose was to compare the performance of prostate-specific membrane antigen (PSMA)-positron emission tomography (PET) traditional fixed threshold (FT) and newly established Prostate Imaging Reporting and Data System (PI-RADS)-based segmented threshold (ST) for diagnosing clinically significant prostate cancer (csPCa). The study retrospectively included 218 patients who underwent multiparametric magnetic resonance imaging (mpMRI) and PSMA-PET examination for suspected prostate cancer (PCa) from January 2018 to November 2021. Lesions with Gleason score ≥ 3 + 4 were diagnosed as csPCa. In PSMA-PET maximum standardized uptake value (SUVmax), the FT for all the lesions and STs for lesions with different PI-RADS score for diagnosing csPCa were determined by receiver operating characteristic (ROC) curves analysis and compared with Z test. The McNemar test was used to compare sensitivity and specificity. Among the 218 patients, there were 113 csPCa and 105 non-csPCa. The PSMA-PET FT was SUVmax > 5.3 (area under the curve, AUC = 0.842) and STs for PI-RADS 3/4/5 were SUVmax > 4.2/5.7/6.0 (AUCs = 0.870/0.867/0.882), respectively. The AUC of PSMA-PET ST was higher than that of PSMA-PET FT (0.872 vs. 0.842), especially for PI-RADS 3 (0.870 vs. 0.653). Multimodality diagnostic criteria combining PSMA-PET ST and PI-RADS scores of mpMRI was established and its AUC was higher than that of PSMA-PET ST (0.893 vs. 0.872) and significantly higher than that of PSMA-PET FT (0.893 vs. 0.842) with an improvement in sensitivity (93% vs. 78%, p < 0.05) without significantly sacrificing specificity (86% vs. 91%, p > 0.05). For diagnosing csPCa, PI-RADS-based PSMA-PET segmented threshold achieved better performance than traditional fixed threshold, especially for PI-RADS 3 lesions. Multimodality diagnostic criteria demonstrated higher diagnostic performance than segmented threshold and significantly better than PSMA-PET fixed threshold for detecting csPCa.


Introduction
Worldwide, prostate cancer (PCa) is one of the most common malignant tumors in men [1].According to Gleason score (GS) [2], PCa can be divided into clinically significant prostate cancer (csPCa, Gleason ≥ 3 + 4) and clinically non-significant prostate cancer (cnsPCa, Gleason = 3 + 3).In the most recent guidelines [3,4], the recommended treatment strategy for cnsPCa is active surveillance, while for csPCa, early intervention is crucial to increase survival.Therefore, the identification of csPCa is a unique clinical need for decision-making.
Multiparametric magnetic resonance imaging (mpMRI), which can provide anatomic structural information and functional information of water molecule diffusion movement simultaneously, has become an integral imaging method for detecting csPCa [5,6].To simplify and standardize the terminology and content of the mpMRI report, the Prostate Imaging Reporting and Data System (PI-RADS) Joint Xiaoli Meng, Wenhui Ma, Weijun Qin, Fei Kang and Jing Wang contributed equally to this work.

Extended author information available on the last page of the article
Steering Committee developed a scale of 1 to 5 based on the degree of suspicion of csPCa [2], with the corresponding csPCa possibility rate ranging from 3 to 81% [7].At present, the most commonly used mpMRI diagnostic criterion for csPCa is PI-RADS ≥ 3 [8,9], but the specificity of this criterion ranges from 37 to 59% [8][9][10].Therefore, increased precision derived from the imaging evaluation would be highly valuable.
Because prostate-specific membrane antigen (PSMA) is overexpressed in more than 90% of PCa [11], 68 Ga-PSMA positron emission tomography-computed tomography (PET/ CT) has become a newly Food and Drug Administration (FDA)-approved csPCa-specific imaging method [12,13].However, for quantitatively determining csPCa, previous studies have established a series of PSMA-PET maximum standardized uptake values (SUVmax) fixed thresholds (FT) ranging from 2.5 to 6.7, with the accuracy/sensitivity/specificity ranging from 77 to 89%/75 to 97%/31 to 84% [14,15], demonstrating widely variable performances and a lack of unified diagnostic standard for PSMA-PET.One possible explanation is that prostatic inflammation mimics the structural changes of csPCa, increasing the background expression level of PSMA, leading to false negative results of fixed thresholds in some low-grade csPCa patients [15,16].Correlation analysis has shown a positive correlation between PI-RADS score and GS of csPCa [17].Therefore, we hypothesized that for lesions with different PI-RADS scores, the PSMA-PET segmented threshold (ST) would be more accurate than fixed threshold (FT) for diagnosing csPCa.
The goal of this study was to establish PSMA-PET segmented thresholds for lesions with different PI-RADS scores and compare the diagnostic performance of segmented threshold and fixed threshold.Furthermore, we endeavored to establish multimodality diagnostic criteria, combining PSMA-PET segmented threshold and PI-RADS scores of mpMRI, to enhance the diagnostic performance of csPCa.

Study Design and Patient Population
The clinical and imaging data of patients with suspected prostate cancer who underwent both PSMA-PET and mpMRI in Xijing Hospital from December 2017 to November 2021 were analyzed retrospectively.The interval between the two examinations was less than 15 days.All patients underwent transrectal ultrasound biopsy and/or radical prostatectomy, and patients for whom the initial biopsy results were negative received at least 6 months of follow-up or repeated histological examination to exclude the possibility of a false negative biopsy.Patients who had already received PCa-related therapy were excluded.According to the histopathological results, the lesions with GS ≥ 3 + 4 were regarded as csPCa, and the lesions with either a negative biopsy or a GS = 3 + 3 were regarded as non-csPCa.The study was approved by the Ethics Committee of Xijing Hospital (approval no.KY20162088-1), and all participating patients provided written informed consent.The research was conducted in adherence with the Declaration of Helsinki and national regulations.

Collection and Evaluation of 68 Ga-PSMA PET/CT Images
Patients underwent PSMA-PET imaging with a Biograph 40 system (Siemens Medical Solutions, Erlangen, Germany). 68Ge/ 68 Ga-generator was from ITG company (Germany), and PSMA-11 ligand was from ABX company (Germany).68 Ga-PSMA-11 was prepared according to a previously published method with a radiochemical purity of > 95% [18].The patients were intravenously injected with 1.8-2.2MBq/ kg body weight 68 Ga-PSMA-11.Whole-body PET scans were acquired approximately 60 min after the tracer injection.The process of image acquisition is consistent with previous reports [18].Positive tumor lesions on PET/CT images were defined as having a higher uptake than the local background and not associated with physiologic uptake, according to the previous literature [19].To calculate SUVmax, a slightly larger iso-contour three-dimensional (3D) spherical volume of interests (VOI) were created in accordance with the location in the prostate with the highest GS by pathology specimen.To keep the VOI of each lesion follow the same segmentation rule, a fixed maximum intensity thresholds (40%) were applied, so that the VOI only includes all voxels equal to or above the chosen percentage of SUVmax, this method has been reported before [20].

Collection and Evaluation of mpMRI Images
All patients underwent pelvic MRI using a GE 3.0T scanner (America) with a 16-channel phased-array coil.The technique and operation parameters of mpMRI were based on previous research [21].Briefly, the scanning sequences included conventional axial T1-weighted imaging (WI), axial T2WI, coronal T2WI, sagittal T2WI, axial diffusionweighted imaging (DWI), and dynamic contrast-enhanced (DCE) images.The apparent diffusion coefficient (ADC) images were reconstructed on AW4.6 post-processing workstation after DWI (barium 0 and 1000 s/mm 2 ) scanning.The PI-RADS scores (version 2.1) were independently evaluated and recorded by two radiologists with 10 and 15 years of experience, respectively, and a joint reading to reach a consensus was performed if there was a dispute.PI-RADS scores were assigned according to the lesions corresponding to the position with the highest GS score in the pathology specimen, and PI-RADS ≥ 3 was defined as a positive result.

Biopsy and Pathology
A transrectal ultrasound-guided 12-core biopsy with necessary additional target biopsy was performed for each patient.Benign diagnoses were defined as repeated negative pathological findings on biopsies or continuous followup by PSA monitoring for at least 6 months [22].Subsequently, a subset of patients with prostate cancer who met the surgical indications underwent radical prostatectomy, and prostatectomy specimens were obtained.The pathological sections were scored according to the consensus of two board-certified genitourinary pathologists to determine their Gleason scores [23,24].Lesions with GS ≥ 3 + 4 were diagnosed as csPCa.The GS was considered to be the highest score on the biopsy or surgery specimens, in case the GS had any contradiction between biopsy and postoperative pathology specimens, only the postoperative pathology specimens would be recorded.

Statistical Analysis
Continuous variables are presented as means ± standard deviations.Categorical variables are presented as frequencies (percentages) and compared with a chi-square test.Inter-rater agreement was evaluated with Cohen's kappa.In PSMA-PET SUVmax, the FT for all the lesions and the STs for lesions of different PI-RADS were, respectively, determined by receiver operating characteristic (ROC) curves and compared with Z test.McNemar tests were used for the head-to-head comparisons of sensitivity and specificity between PSMA-PET FT, STs, and multimodality diagnostic criteria.The diagnostic performance of different diagnostic criteria was compared by using the area under the curve (AUC), sensitivity, and specificity.The data were analyzed by MedCalc (version 19.0.5, MedCalc Software), Graph-Pad Prism software (version 6.0, GraphPad Software), and R software (version 3.4.2,R Foundation for Statistical Computing).All hypothesis tests were two-sided, and p < 0.05 was considered to indicate statistical significance.

Patient Characteristics
There were 324 candidate patients with clinically suspected prostate cancer who underwent mpMRI and PSMA PET/ CT examination, and after applying inclusion and exclusion criteria, 218 patients were enrolled in this study, including 113 in the csPCa group and 105 in the non-csPCa group (which included 13 cases of cnsPCa and 92 cases of benign prostatic hyperplasia).There were 150 patients who underwent biopsy only without subsequent surgery, based on their GS.Sixty-eight patients underwent biopsy followed by surgery, and five of the 68 patients had GS upgraded from 6 to 7 following surgery.
The kappa value of PI-RADS scores recorded by between the two radiologists was 0.82; there were 20 cases (9.2%) with initially inconsistent results that required a joint reading.
As shown in Table 2, for lesions with PI-RADS 3, the AUC values of mpMRI and FT were significantly lower than ST (0.617 vs. 0.870, p < 0.001; 0.653 vs. 0.870, p < 0.001) (Fig. 1a and e).For lesions with PI-RADS 4, the AUC values of ST and FT were higher than mpMRI (0.862 vs. 0.768, p < 0.05; 0.867 vs. 0.768, p < 0.05).The AUC of ST was slightly higher than that of FT, but there was no significant difference (p > 0.05) (Fig. 1b and e).For PI-RADS 5, the AUC of mpMRI was higher but did not reach a statistical difference compared with FT and ST (0.903 vs. 0.871, p > 0.05; 0.903 vs. 0.882, p > 0.05) (Fig. 1c and e).To sum up, the diagnostic performances of PSMA-PET segmented thresholds were higher than that of mpMRI in PI-RADS scores of 3 and 4, especially for PI-RADS 3 lesions, whereas, for a PI-RADS score of 5, the diagnostic performance of mpMRI was higher than that of PSMA-PET segmented threshold.
For the overall patient population, the AUC of mpMRI was lower than the PSMA-PET FT and ST (0.630 vs. 0.842, p < 0.001; 0.630 vs. 0.872, p < 0.001) (Fig. 1d and  e).The main improvement in the diagnostic performance of PSMA-PET compared with mpMRI was the specificity, especially in PI-RADS 3 lesions (37.0% to 98.8% for FT and 37.0% to 87.7% for ST, respectively).Compared to FT, the main advantage of ST was avoiding a missed diagnosis of csPCa with low malignant risk by mpMRI, as the sensitivity for the diagnosis of csPCa increased from 31.8 for FT to 86.4% for ST.

Establishment of Multimodality Diagnostic Criteria and Performance Comparison
According to the above-mentioned differences in performance of thresholds for diagnosing csPCa, multimodality diagnostic criteria combining PSMA-PET ST and PI-RADS scores of mpMRI were established: PI-RADS 3 and SUVmax > 4.2, or PI-RADS 4 and SUVmax > 5.7, or PI-RADS 5 (Fig. 2).The AUC of the established multimodality criteria was higher than that of PSMA-PET segmented threshold (ST) (0.893 vs. 0.872) and significantly better than PSMA-PET fixed threshold (FT) (0.893 vs 0.842) with an improvement in sensitivity (93% vs. 78%, p < 0.01) without significantly sacrificing specificity (86% vs. 91%, p > 0.05) (Fig. 3a and Table 3).The clinical decision curve analysis (DCA) showed that the net benefit of multimodality diagnostic criteria for detecting csPCa was the highest when the threshold probability was less than 0.7 (Fig. 3b).
As shown in Table 3, the main advantage of the multimodality diagnostic criteria compared with PSMA-PET ST is its higher sensitivity (93% vs. 85%, p < 0.01).When the PI-RADS score was low, the sensitivity was improved based on ST which avoided a missed diagnosis.Typical cases are shown in Fig. 4.Moreover, lesions with low expression of PSMA could avoid being omitted when the PI-RADS score was 5. Illustrative cases are shown in Fig. 5.

Discussion
Our study had the following principal findings, firstly, for lesions with different PI-RADS scores, corresponding PSMA-PET segmented threshold could be calculated, and the diagnostic performances of segmented threshold were superior to that of fixed threshold.Secondly, the diagnostic performances of PSMA-PET segmented thresholds were higher than that of mpMRI in PI-RADS scores of 3 and 4, whereas for a PI-RADS score of 5, the diagnostic performance of mpMRI was higher than that of PSMA-PET segmented threshold.Thirdly, multimodality diagnostic criteria demonstrated higher diagnostic performance than segmented threshold and significantly better than PSMA-PET fixed threshold for detecting csPCa.
Regarding the diagnostic threshold of PSMA-PET, this study reinforced the necessity of establishing different segmented thresholds and validated the superiority of segmented thresholds compared to fixed threshold for detecting csPCa.At present, the SUVmax is the main reference index for PSMA-PET diagnosis, but there is no unified quantitative standard.Previous studies have established a series of PSMA-PET fixed thresholds for detecting csPCa ranging from 2.5 to 6.7 [13][14][15].For example, a study by Marel D established a PSMA-PET fixed threshold (SUVmax ≥ 2.5) and found that PSMA-PET improved the specificity for csPCa compared with mpMRI [13].However, prostatic inflammation can simulate the structural changes of PCa and increase the background expression level of PSMA.Therefore, a fixed threshold may lead to false negative results in some low-grade csPCa patients [25,26].This study proposed using PSMA-PET segmented threshold of different malignant risk lesions, with its main advantage that it significantly improved the sensitivity of PI-RADS 3 lesions compared with fixed threshold, indicating that PSMA-PET segmented threshold could avoid some missed diagnoses of low-grade csPCa.This verified the necessity that for lesions with varying degrees of malignancy, they should be classified by different PI-RADS scores to establish different PSMA-PET segmented thresholds for detecting csPCa in clinical practice.
Our study found that the diagnostic performances of PSMA-PET segmented thresholds were higher than that of mpMRI in PI-RADS 3 and 4. By contrast, in PI-RADS 5, mpMRI was superior to PSMA-PET segmented threshold.It is important to note that there are csPCa and non-csPCa lesions that have been reported to show overlapping mpMRI characteristics [27][28][29].They presented as ill-defined homogenous isointense or hypointense lesions on T2WI, usually multifocal, with moderate to low ADC values and moderate signal intensity on high-b-value images.In the mpMRI PI-RADS scoring system, such an anomaly is assigned with an overall assessment score 3 and 4 [30,31], especially in PI-RADS 3, indicating that the presence of a csPCa is indeterminate.However, lesions with a PI-RADS score of 5 were usually easy to discriminate from non-csPCa lesions especially if the presentation was a compact and dense lesion located in the peripheral zone (PZ) [31,32].In the PZ the majority of PI-RADS, 5 lesions were easily recognized on T2WI due to their low or very low signal  intensities with a well-defined but irregular area.Fibromuscular stroma and fibromuscular hyperplasia, however, show similar very low signal intensities and may be confounders of PI-RADS 5 lesions on T2WI [33].While on high-bvalue DWI images fibromuscular stroma and fibromuscular hyperplasia show moderate signal intensity, PI-RADS 5 lesions show high signal intensity, facilitating differentiation between these lesions.Therefore, multimodality diagnostic criteria combined the advantages of sensitivity of mpMRI and specificity of PSMA-PET in the diagnosis of csPCa.Similar to the PI-RADS score, Rowe et al. recently developed the PSMA-RADS 1.0 grading diagnostic standard of prostate PSMA PET/CT [13].Furthermore, the clinical availability of scanners for integrated PET and mpMRI has enabled the potential for multimodal, anatomical, and functional imaging to be explored [34][35][36].The multimodal quantitative diagnosis standard of PET + MRI demonstrated in this study reflects the complementary value of PET and MRI in the diagnosis of csPCa and provides a reasonable basis for the clinical application of co-registration and integration PET/MRI.Many previous studies have emphasized the comparison of advantages and disadvantages of PSMA-PET versus MRI [37][38][39], overlooking the potential of the two studies in combination.To emphasize, mpMRI has a sensitivity advantage in malignant lesions with negative expression of PSMA, and PSMA-PET has a specificity advantage in inflammatory lesions with positive mpMRI.This study established multimodal diagnostic criteria for the detection of clinically significant prostate cancer, demonstrating the superiority of PET + MRI.If verified by larger clinical studies with potential refinement of threshold standards, these objective and immediately operable diagnostic criteria will be more conducive to acceptance and endorsement of PET + MRI multimodal diagnostic technology to improve the diagnostic accuracy of csPCa.
This study has several limitations.First, although this research confirmed the importance of PSMA-PET STs, variations could have existed due to differences in scanner calibration [40,41].Therefore, SUVmax thresholds may not yet be generalizable across sites, and larger multicentre studies using different types of PET/CT scanners may be required to validate the reproducibility.Second, only two parameters, SUVmax of PSMA-PET and PI-RADS score of mpMRI, were included in our study.Other parameters such as the mapping value of mpMRI, the ADC value, the pattern of PSMA, and the characteristics of artificial intelligence have also been shown to be diagnostically efficient in recent studies.Further research is needed to confirm whether these parameters have the potential to be added to the multimodality diagnosis criteria [5,42].Third, studies that have assessed cost-effectiveness have become hot topics in recent years; whether multimodal imaging of csPCa improves performance while maintaining reasonable cost requires further investigation.

Conclusions
PI-RADS-based segmented threshold achieved better performance than fixed threshold for detecting csPCa, especially for PI-RADS 3. The multimodality diagnostic criteria combining PSMA-PET ST and mpMRI demonstrated higher diagnostic performance than single-modality criteria.

Fig. 1 aFig. 2
Fig. 1 a-c The ROC curves of mpMRI (blue line), PSMA-PET FT (green line), and PSMA-PET ST (red line) for detecting csPCa in PI-RADS 3/4/5.d The ROC curve of the three criteria for detecting csPCa in the population.e The bar chart of the area under the ROC

Fig. 3 a
Fig. 3 a The bar chart of the sensitivity, specificity, and AUC of the PSMA-PET FT, ST, and multimodality diagnostic criterion for detecting csPCa.b The clinical decision curve analysis (DCA); when the threshold probability is less than 0.7, the net benefit of multimo-

Fig. 4 AFig. 5 A
Fig.4A 66-year-old male with total serum PSA level of 9.9 ng/mL.a Small patches of increased radioactivity uptake in the prostate can be seen on the PET MIP map.mpMRI showed b patchy low signal intensity of T2WI in the right transitional zone of the prostate, c with unclear boundary, slightly high signal intensity on DWI, and d slightly low signal intensity on ADC, with a PI-RADS score of 3.

Table 1
Characteristics of patients with non csPCa and csPCa csPCa clinically significant prostate cancer, PSA prostate-specific antigen, PI-RADS Prostate Imaging Reporting and Data System

Table 3
Comparison of the performance of PSMA-PET fixed threshold, segmented threshold, and multimodality diagnostic criteria for detecting csPCa Sensitivity and specificity p values were determined from McNemar test; AUC p value was determined from Z test