Accuracy of standard clinical 3T prostate MRI for pelvic lymph node staging: Comparison to68Ga-PSMA PET-CT

The aim was to assess the performance of prostate 3T MRI for pelvic lymph node (LN) staging in prostate cancer (PCa), in comparison to 68Gallium-prostate specific membrane antigen PET-CT (68Ga-PSMA PET-CT) as reference standard for LN detection. 130 patients with PCa underwent non-contrast-enhanced multiparametric prostate 3T MRI and 68Ga-PSMA-PET-CT within 180 days at our institution. Overall, 187 LN metastases (n = 43 patients) detected by 68Ga-PSMA-PET-CT were characterized by calculating maximum standardized uptake value (SUVmax), area, diameter and anatomical location including iliac, obturator, presacral and inguinal region. MRI achieved an overall sensitivity, specificity, positive and negative predictive value of 81.6% (CI 71.1–88.9%), 98.6% (CI 97.6–99.2%), 73.5% (CI 52.1–87.6%) and 99.5% (CI 98.8–99.8%), respectively. On a region-based analysis, detection rates differed non-significantly (ps > 0.12) in the anatomical regions. On a size-dependent analysis, detection of LN > 10 mm did not differ significantly (ps > 0.09) from LN ≤ 10 mm. In comparison to single T1 sequence evaluation, additional use of the T2 weighted sequences did not improve the overall performance significantly (p > 0.05). 3T prostate MRI represented an accurate tool for the detection of LN compared to 68Ga-PSMA-PET-CT. Especially for LN metastases smaller than 10 mm, MRI was less accurate compared to 68Ga-PSMA-PET-CT.


Detection of lymph node metastases in MRI dependent on anatomical regions.
In the analysis dependent on anatomical regions the sensitivities differed between 66.8% and 100% while specificities ranged between 97.1% and 100%. The highest detection rate was presented in the inguinal region with a 100% (CI 0-100%) followed by the obturator right and left region with 88.8% (CI 75.0-95.5%) and 84.9% (CI 62.8-95.0%). Presacral LN were detected with a sensitivity was 82.0% (CI 50.0-95.4%). The lowest sensitivity was presented in the iliac region with 80.7% (CI 60.0-92.1%) for the left and 66.8% (CI 46.6-82.2%) for the right region. Next, we evaluated possible differences in MRI diagnostical accuracy in the different anatomical regions. The six regions had a significant additional impact on diagnostical accuracy beyond the MRI judgments (χ²(10) = 18.7, p < 0.05). However, the source of this impact could not be pinpointed, as all single interaction effects were non-significant  Table 1. Characteristics of all investigated lymph nodes in 68 Gallium-Prostate specific membrane antigen PET-CT. This table presents the main characteristics of all lymph nodes (LN) depicted in 68 Gallium-Prostate specific membrane antigen PET-CT including all benign and malignant LN. Count, area, long-axis diameter, short-axis diameter and size-ratio of the LN are presented in the columns. The size ratio is described as quotient of short-axis diameter divided through long-axis diameter. The rows show all LN together and fielded in the six defined anatomical regions iliac left and right, obturator left and right, presacral and inguinal region. Data are given in means and standard deviations. Abbreviations: LN = Lymph nodes, LAdm = Long-axis diameter, SAdm = Short-axis diameter, SD = Standard deviation. www.nature.com/scientificreports www.nature.com/scientificreports/ in logistic regression (ps > 0.12). Differences in the diagnostical accuracy were calculated relative to the iliac left region, which was arbitrary and was non-significant as can be seen at the overlapping CI of the sensitivities and specificities throughout the regions. Please refer to Table 3 for further details.
Assessment of histopathological data compared to MRI sensitivity. To assess a possible impact of histopathologic data to the readings, an interaction test was performed in this study. The diagnostical accuracy of MRI was significantly influenced by the results of the biopsy (χ²(1) = 4.8, p < 0.05). Sensitivity increases with greater Gleason scores from biopsy, i.e., p = 0.21 for a Gleason score of 6 and p = 0.97 for a score of 10. Gleason scores at the time of biopsy ranged from 6 to 10 with a mean of 7.8. Due to missing final Gleason scores after prostatectomy in most patients, no reliable statements could be made apart from biopsy. Impact of the additional T2 and DWI sequences in lymph node staging. Since LN detection in MRI is commonly related to T1 sequence evaluation, we assessed the performance of T1 plus T2 and DWI compared to single T1 sequence evaluation. 130 LN were detected in the single T1 evaluation presenting an overall sensitivity, specificity, PPV and NPV of 71.8% (CI 58.2-82.3%), 99.0% (CI 98.3-99.5%), 72.2% (CI 47.9-88.0%) and 99.5% (CI 98.5-99.8%). Like T1 + T2, single T1 sequence evaluation displayed substantial diagnostical accuracy. Mean area was 1.0 ± 1.2 cm 2 (range 0.1-9.1 cm 2 ) and a mean size ratio of 0.8 ± 0.2 (range 0.3-1.) in MRI. False negative LN had a mean area, size ratio and SUVmax of 0.5 ± 0.3 cm 2 (range 0.1-1.8 cm 2 ), 0.7 ± 0.2 (range 0.3-1.5) and 6.4 ± 4.0 (range 0.7-18.9) in 68 Ga-PSMA-PET-CT. False positive LN had a mean area of 1.2 ± 0.7 cm 2 (range 0.3-3.0 cm 2 ) and a size ratio of 0.7 ± 0.2 (range 0.4-1.0) in MRI. The sensitivities in single T1 sequence evaluation differed between 65.0% and 100% while specificities ranged between 97.7% and 100%. The highest detection rate was presented in the inguinal region with a 100% (CI 0-100%) followed by the presacral region with 82.3% (48.8-95.8%). LN in the obturator left region were detected with a sensitivity of 75.9% (CI 50.0-91.0%) followed by iliac left with 73.1% (CI 48.9-88.5%) and the obturator right region with 67.5% (CI 47.2-83.0%). The lowest sensitivity was seen in the iliac right region with 65.0% (CI 43.2-81.9%). Although the sensitivity from T1 alone was lower than combined with T2 + DWI, this difference was not significant, as can be seen from the overlapping CIs between single T1 versus T1 combined with T2 and DWI sequence evaluation. Please refer to Table 4 and

Discussion
This study demonstrated that high resolution 3 T prostate MRI represents an accurate tool for the detection of LN metastases. Especially for LN metastases smaller than 10 mm, MRI was less accurate compared to 68 Ga-PSMA-PET-CT. In the region-based analysis, the performance of MRI was not significantly (p > 0.05) different throughout the anatomical regions. The highest detection rate for MRI was achieved for the inguinal, obturator and presacral region, while the lowest sensitivities were achieved in the iliac regions. This may be owed to the challenging anatomical conditions and flow artifacts resulting from these vessels. Detection rates were higher in areas where LN can be better delineated from surrounding structures. Figures 2 and 3 present examples of direct and challenging detection of LN in PET-CT and MRI. In comparison to single T1 sequence evaluation, T1 combined with T2 and DWI sequence evaluation showed no statistically significant (p > 0.05) higher overall sensitivity and in the region-based analysis, underlining the relevance of sole T1 sequence evaluation. An example of additional T2 sequence evaluation is presented in Fig. 4.
On a size-based analysis, LN ≤ 10 mm in the iliac, obturator and presacral region were frequently missed, whereas LN ≤ 10 mm were detected more reliable in all anatomical regions but no statistically significant (p > 0.05) lower detection rate was seen for LN ≤ 10 mm compared to LN > 10 mm. Similar non-significant  www.nature.com/scientificreports www.nature.com/scientificreports/ (p > 0.05) effects were found for single T1 sequence evaluation showing no significant additional impact of T2 evaluation based on LN size. For additional details please refer to the supplementary dataset. 68 Ga-PSMA-HBEDD-CC is an inhibitor of the glutamate carboxypeptidase II, labelled with 68 Gallium 21 . PSMA is a cell-surface transmembrane protein found in the prostate, brain, lacrimal and salivary glands, tumor neovasculature, tubules of the kidney and intestine 22 . It is a 110 kDa highly glycolysated peptidase and belongs to a family of zinc-dependent exopeptidases with glutamate carboxypeptidase activity 22,23 . It is highly active in prostatic intraepithelial neoplasia and metastatic PCa 22 . As of today, only few numbers of studies assessed the performance of 68 Ga-PSMA PET-CT for LN staging in PCa patients and the diagnostic accuracy was tested regularly against histopathology as standard reference. Recently, Hamed et al. presented a prospective study with 106 out of 165 patients presenting with local recurrence and extraprostatic metastases, in which sensitivity, specificity and accuracy of 68 Ga-PSMA was 99.0%, 100% and 98.8%, respectively, compared to histopathology 24 . In 2016, a study was published by Herlemann et al. on 34 patients undergoing a 68 Ga-PSMA PET-CT prior to pelvic lymph node dissection (PLND) reporting an overall sensitivity, specificity, NPV and PPV of 84.0%, 82.0%, 84.0% and 82.0%, respectively 25 . Subsequent studies also demonstrated favorable detection rates for 68 Ga-PSMA PET-CT in comparison to histopathology 26,27 . These results underline the value of this imaging technique, while PLND is an invasive procedure associated with perioperative risks such as lymphedema and venous thromboembolism 28 . Another major limitation of histopathological assessment is the sampling error 8 . Skip metastases near the common and internal iliac vessels can be missed due to the limited exploration in the surgery 8 .
Histopathological evaluation of LN is mainly dependent on morphologic criteria such as enlarged diameter or rounded LN or an increased LN volume 29 . Another input is set through the extranodal extension of LN metastases which is defined as a perforation of the LN capsule resulting in an expansion into extranodal tissues 30 . In 1998 Cheng et al. published a study including 269 patients, with LN metastases, presenting a significant correlation between nodal cancer volume and Gleason score recommending, that the diameter of the largest LN should be evaluated as prognostic factor of progression to distant metastasis rather than the number of LN 30 .In a follow-up       www.nature.com/scientificreports www.nature.com/scientificreports/ extranodal extension resulting in an elevated risk for patients with extranodal extension in PLND to develop a biochemical recurrence or distant metastasis 32 .
Today, MRI of the prostate is a clinical routine and highly sensitive imaging procedure for staging of patients and detection of extracapsular and seminal vesicle infiltration due to its excellent anatomical resolution 11,12 . The performance for LN staging is still considered to be challenging 12 . A meta-analysis of Hövels et al. included 24 studies with a mean sensitivity and specificity of 39.0% and 82.0% with ranges of sensitivities and specificities of 6.0-83.0% and 65.0-99.0%, respectively for LN detection by MRI prior to PLND while magnetic field strength however remained unknown for all investigated studies 5 . Sensitivity, specificity, PPV and NPV of 71.4%, 94.7%, 62.5% and 96.4%, respectively, were presented in a study by Kim et al. in 2010 for non-contrast enhanced T1 and T2 sequences of 1.5T MRI using surface coils for LN staging in comparison to histopathology 6 . In 2017, Gupta et al. reported sensitivity, specificity, PPV and NPV of 25.9%, 98.6%, 70.0% and 91.4%, respectively for LN detection using non contrast enhanced 1.5T MRI 9 .
Since area, LAdm, SAdm and size ratio are major signs for malignancy in MRI, high resolution 3T MRI showed an improved detection rate of small LN ≤ 10 mm in our study. Interestingly, a generally lower threshold for malignancy (LAdm of <10 mm) presented a better performance in comparison to histopathology than a higher threshold in a previous study, which is comparable to our findings 5 . In 2016, Barchetti et al. compared the performance of 1.5T MRI with non-contrast enhanced T1 plus T2 sequences in comparison to 18 F-Choline PET-CT exams in 152 patients with biochemical recurrence reporting a sensitivity, specificity, PPV and NPV of 98.0%, 99.0%, 97.0% and 98.0%, respectively 10 . This good performance of 1.5T MRI has to be seen in the context of the chosen reference standard. Tulsyan et al. examined the usefulness of 68 Ga-PSMA PET-CT in 36 patients with a biopsy proven PCa with a minimum Gleason score of 8 and PSA blood levels >20 ng/ml for LN staging in comparison to non-contrast enhanced 1.5T MRI 11 . 29 out of 36 patients presented LN in 68 Ga-PSMA PET-CT in contrast to 20 MRI positive patients, which resulted in a concordance of 72% between both modalities without sensitivities or specificities given in the manuscript 11 .
Regarding the use of 3T MRI, Zattoni et al. published a study on recurrent PCa after failure of primary radiation therapy using histopathology as reference standard for LN detection by contrast enhanced 3T MRI applying an endorectal coil. They reported a sensitivity and specificity of 60.0% and 85.7%, respectively, for an unknown number of LN 15 . A further study from Zhang et al. compared contrast enhanced multiparametric MRI at 3T and 68 Ga-PSMA PET-CT against histopathology for a cohort of 42 patients prior undergoing radical prostatectomy with PLND and found to be equal regarding the diagnostic accuracy 14 14 . In our study 68 Ga-PSMA-PET-CT was set as reference standard with an overall sensitivity, specificity, PPV and NPV of 81.6%, 98.6%, 73.5% and 99.5% for MRI. In contrast to our study, Zhang et al. did not present LN detection rates according to anatomical regions while showing a higher overall detection rates with comparable results in the size-based analysis compared to our study. This may be due to the rather small cohort of 42 patients and the elevated mean PSA blood level with 52.3 ng/ml in Zhang et al. compared to 15.8 ng/ml in our study. Furthermore, the delay between PET-CT, MRI and surgery remained unknown. Moreover, the used malignancy criteria for MRI in Zhang et al. were SAdm > 10 mm, a rounded LN with a SAdm > 8 mm, increased contrast enhancement or a diffusion restriction in DWI and ADC map 14 . Our study chose a stricter malignancy criteria with a LAdm excess of 10 mm or a rounded LN defined through the quotient of SAdm divided through LAdm was present. In contrast to Zhang et al., the investigated MRI sequences in this study did not include the examination of DCE sequences and no use of contrast agent in our study, since gadolinium is associated with certain risks including nephrogenic systemic fibrosis especially in patients with chronic kidney disease 33 .
In summary, higher PSA blood levels may indicate more advanced tumor stages, the unknown delay of the procedures, the additional use of DCE sequences and contrast agent may have resulted in the slightly higher detection rates in Zhang et al. compared to our study. The comparable results of our study underline the validity of 68 Ga-PSMA-PET-CT as reference standard compared to histopathology.
Limitations are the retrospective character of this study and the relatively small number of patients involved. No histopathological confirmation of the metastases seen in 68 Ga-PSMA PET-CT was performed. A delay of up to 180 days between both imaging techniques might have influenced the size of the LN, but because PCa is a slow growing cancer entity, this time delay was considered of minor relevance. The performance of MRI detection was shortened through challenging localization near big vessels, and therefore interobserver variability cannot be excluded. Pitfalls of physiological tracer uptake were limited through parallel evaluation in CT. Size variations or necrosis of LN may have biased the examination as well. Given that size is a determining factor of malignancy in MRI, the size-based analysis could be biased leading to better sensitivities for LN > 10 mm compared to LN ≤ 10 mm. To limit this effect, the examination was done by two experienced readers and a size ratio was included to detect small but malignant LN based on a previous study in T1 sequences 29 .
In conclusion, high resolution 3T prostate MRI represents an accurate tool for the detection of LN metastases. The detection rates of MRI were lower for metastases in complex anatomical regions, compared to 68 Ga-PSMA PET-CT. Especially for LN metastases smaller than 10 mm, MRI was less accurate compared to 68 Ga-PSMA-PET-CT. The authors suggest that 68 Ga-PSMA-PET-CT should be used for primary lymph node staging and for patients with biochemical recurrence.  0-180 days). The mean delay between diagnosis and the first examined scan (PET-CT or MRI) in this study was 1.7 ± 3.4 years (range 0-17 years) with a median of 0 years. When multiple studies were present, the scans with the shortest delay were used in this investigation. All other examinations were not used as part of the reading. Patients in this study did not receive any surgery, radiotherapy, change in the regime of chemotherapy or change of hormonal treatment within the delay between both modalities. There was no artificial delay of treatment in the patients investigated in this study. Prostate specific antigen (PSA) blood levels were 15.8 ± 23.6 ng/ml (range 0.2-196 ng/ml) collected within 26.5 ± 43.0 days (range 0-193 days) to the 68 Ga-PSMA PET-CT and a median core needle biopsy Gleason score of 8 (range 6-10) was reported. Patients characteristics are presented in Table 5.
68 Ga-PSMA PET-CT and 3T MRI acquisition protocols. A standard 68 Ge/ 68 Ga generator (Eckert and Ziegler) was used for elution of 68 Ga prior to labelling with PSMA-HBED-CC (ABX GmbH, Radeberg, Germany) 22,23,34 . After injection of 129.0 ± 26.2MBq of 68 Ga-PSMA-HBED-CC, a low dose CT for attenuation correction (120 kVp, 30 mAs) and anatomical mapping was acquired within 89.0 ± 42.5 min immediately before the PET scan, using a Gemini TF 16 Astonish PET-CT scanner (Philips medical systems) 35 . All 130 patients underwent a non-contrast-enhanced prostate multiparametric MRI at our institution at 3T (Magentom Skyra, Siemens Healthcare, Erlangen, Germany). Standard prostate MRI acquisition protocol included high resolution T2 weighted high resolution turbo spin echo sequences (T2 HR TSE,  Image analysis. Visage 7.1 (Visage Imaging) was used as the standard software package. Low dose whole body CT sequences and 68 Ga-PSMA PET sequences were automatically fused for the evaluation process. Since the acquisition of a MRI of the pelvis is limited to a localized area while 68 Ga-PSMA PET-CT covered the whole body, the aortic bifurcation was set as the upper border for positive LN in 68 Ga-PSMA PET-CT. Consensus reading was performed by two readers. All images were analysed independently in a blinded and random order.

Assessment of lymph node metastasis in MRI.
For LN examination in 3T MRI data sets, T1 TSE, T2 TSE and DWI sequences were used. T1 sequences were used in the first step, followed by high resolution T2 TSE and DWI to assess their additional value. The diameter of suspicious LN were measured in axial planes through manual delineation of a ROI, resulting in size, long-axis diameter (LAdm) and short axis diameter (SAdm). SAdm was defined as the rectangular line of the LAdm. A LN was defined as positive in MRI when the SAdm divided through LAdm exceeded the size ratio of 0.8 or if the LAdm was ≥10 mm, based on a previous study on LN detection in T1 sequences 29 . Signal intensity was no criteria for definition of a LN as metastatic. The size ratio is reported throughout the manuscript. All LN were characterized in six levels according to the adjacent anatomical structures dividing them into presacral, inguinal LN and alongside the large arterial vessels. LN alongside the Gleason score after surgery 7.7 0.9 7 7-9 Table 5. Basic characteristics of study collective. The basic characteristics of the prostate cancer patients who received a 68 Gallium-Prostate specific membrane antigen ( 68 Ga-PSMA) PET-CT and a multiparametric MRI within 180 days investigated in this study are presented in this table. This included the age of the patients, the delay between both imaging modalities, the prostate specific antigen blood level, the delay towards the 68 Ga-PSMA PET-CT, prostate volumes, the Gleason scores from biopsy and after surgery and the tumor stages. Data are given in means, standard deviations, medians and ranges. Abbreviations: PSMA = Prostate specific membrane antigen, PET = Positron emission tomography, MRI = Magnetic resonance imaging, PSA = Prostate specific antigen blood level.
www.nature.com/scientificreports www.nature.com/scientificreports/ common and external iliac arteries were characterized as the iliac right and iliac left region as superordinate levels and LN near the internal iliac and obturator artery were characterized as obturator right and obturator left region. Please refer to Fig. 5 for visualization of the defined regions.

Measurement of lymph node metastasis in 68
Ga-psMA pet-Ct. All 68 Ga-PSMA avid LN were measured in the axial plane at maximum diameter using the CT sequence for delineation of the ROI after evaluation of tracer uptake in 68 Ga-PSMA PET overlay at isocontour of 50%. LN were defined positive, when an abnormal focal tracer signal with a higher signal intensity than the surrounding background was detected in 68 Ga-PSMA PET and a LN in CT could be allocated to the signal 14 . LN size evaluation did not affect the definition of positivity in 68 Ga-PSMA PET-CT, which is comparable to a previously published study 14 . In addition, maximum standardized uptake values (SUVmax) were assessed. All visible LN on CT without tracer uptake were measured in LAdm and SAdm to be defined as true negatives. All LN depicted in PET-CT were divided into the 2 subgroups LN > 10 mm and LN ≤ 10 mm. MRI sensitivity was calculated for both size groups and for each anatomical region. Please refer to Fig. 6 for an example of a LN < 10 mm and Fig. 7 for a LN > 10 mm using PET-CT and MRI. statistical analysis. 68 Ga-PSMA PET-CT was set as reference standard in this study.
Descriptive statistics were done using MedCalc Statistical Software version 17.6 (MedCalc Software bvba; http://www.medcalc.org; 2017) and R software (Version 3.5.0, Vienna, Austria, https://www.R-project.org, +lme4-package) was used for multi-level logistic regression. We used logistic regression in order to assess the diagnostic quality of MRI in contrast to our gold standard. The overall fit of a logistic regression model corresponds to the overall predictive accuracy and is as such related to comparable analysis techniques like ROC analysis. We utilized likelihood-ratio chi-square tests to calculate p-values within logistic regression. More specifically, we applied multi-level logistic regression to satisfy our hierarchical data structure (lymph nodes within patients) Multi-level regression adjusts for clusterings in data, i.e., for effects that some patients a conspicuous lymph  Example of a lymph node smaller than 10 mm in 68 Gallium-Prostate specific membrane antigen PET-CT compared to T1 MRI. This figure shows a suspicious lymph node smaller than 10 mm diameter. The lymph node is visualized in corresponding axial plane slices using CT, 68 Gallium-Prostate specific membrane antigen PET-CT and T1 MRI from left to right. The suspected lymph node is located at the right common iliac artery and is challenging to detect in T1 MRI. The lymph node is highlighted through red arrows. (A) Lymph node depicted in CT, (B) Lymph node depicted in 68 Gallium-Prostate specific membrane antigen PET-CT, (C) Lymph node depicted in the T1 sequence in MRI.
www.nature.com/scientificreports www.nature.com/scientificreports/ node might be more likely to have more conspicuous lymph nodes 36 . Multi-level models include such effects and can even capture person-wise differences in diagnostic accuracy (called random slope models). In our analyses below, we tested for each predictive model whether such person-wise differences were statistically significant. The common properties of a diagnostical test (sensitivity, specificity, positive predictive value and negative predictive value) were calculated from logistic regression and confidence intervals (CI 95%) were given through logistic regression using the method proposed by Coughlin et al. 37 . When no confidence intervals could be calculated due to perfect agreement of MRI and PET-CT, it is highlighted in the tables. A p-value p < 0.05 was considered statistically significant. Figure 7. Example of a lymph node larger than 10 mm in 68 Gallium-Prostate specific membrane antigen PET-CT compared to T1 MRI. This figure shows a suspicious lymph node larger than 10 mm diameter. The lymph node is visualized in corresponding axial plane slices using CT, 68 Gallium-Prostate specific membrane antigen PET-CT and T1 MRI from left to right. The lymph node is located at the right external artery. The lymph node is highlighted through red arrows. (A) Lymph node depicted in CT, (B) Lymph node depicted in 68 Gallium-Prostate specific membrane antigen PET-CT, (C) Lymph node depicted in the T1 sequence in MRI.