Subnodal Correspondence of PSMA Expression and USPIO-MRI in Metastatic Pelvic Lymph Nodes in Prostate Cancer

Objectives Two advanced imaging modalities used to detect lymph node (LN) metastases in prostate cancer patients are prostate-specific membrane antigen (PSMA) positron emission tomography/computed tomography and ultrasmall superparamagnetic iron oxide (USPIO)–enhanced magnetic resonance imaging (MRI). As these modalities use different targets, a subnodal comparison is needed to interpret both their correspondence and their differences. The aim of this explorative study was to compare ex vivo 111In-PSMA μSPECT images with high-resolution 7 T USPIO μMR images and histopathology of resected LN specimens from prostate cancer patients to assess the degree of correspondence at subnodal level. Materials and Methods Twenty primary prostate cancer patients who underwent pelvic LN dissection were included and received USPIO contrast and 111In-PSMA. A total of 41 LNs of interest (LNOIs) were selected for ex vivo imaging based on γ-probe detection or palpation. μSPECT and μMRI acquisition were performed immediately after resection. Overlay of μSPECT images on MR images was performed, and the level of correspondence (LoC) between μSPECT and μMR findings was assessed according to a 4-point Likert classification scheme. Results Forty-one LNOIs could be matched to an LN on ex vivo μMRI. Coregistration of μSPECT and USPIO-enhanced water-selective multigradient echo MR images was successful for all 41 LNOIs. Ninety percent of the lesions showed excellent correspondence regarding the presence of metastatic tissue and affected subnodal site (LoC 4; 37/41). In only 1 of 41 LNOIs, a small metastasis was misclassified by both techniques. Three LNOIs were classified as LoC 3 (7%) and 1 LNOI as LoC 2. All LoC 2 and LoC 3 lesions had PSMA-expressing metastases on final histopathology Conclusions Coregistration of μSPECT and USPIO-μMRI showed excellent subnodal correspondence in the majority (90%) of LNs. Ex vivo imaging may thus help localize small cancer deposits within resected LNs and could contribute to improved interpretation of in vivo imaging of LNs.

I n patients with prostate cancer with intermediate-or high-risk dis- ease, lymph node (LN) metastases often first occur in the pelvis or lower abdomen.Nodal staging is crucial in determining prognosis and treatment options. 1 Although extended pelvic LN dissection (ePLND) is the criterion standard for assessing nodal status, prostate-specific membrane antigen (PSMA) positron emission tomography/computed tomography PET/CT is increasingly used as a noninvasive imaging modality to identify and localize metastatic sites before ePLND. 2 Recent data, however, show limited sensitivity of PSMA-PET/CTwith histopathology as the reference standard.0][11] This technique uses dextran-coated iron oxide nanoparticles (ferumoxtran-10), which are administered intravenously and accumulate only in healthy lymphatic tissue over 24-36 hours through phagocytosis by macrophages. 12,13Sensitive for the presence of these accumulated nanoparticles, USPIO-enhanced MRI can discriminate between healthy LNs and LNs that are (partially) infiltrated by metastatic tumor cells. 14,15Routine 3-dimensional (3D) T2*-weighted sequences with submillimeter isotropic resolution can visualize LNs as small as 2 mm. 16he targets on which imagings with PSMA-PET/CT and USPIO-MRI are based are very different; PSMA-PET targets PSMA expression on tumor tissue, whereas USPIO-MRI is a more anatomical imaging modality, depending on signal retention in tumor tissue as nanoparticles cause signal loss in healthy lymphatic tissue.Consequently, findings of the 2 imaging techniques may be complementary.A study comparing PSMA-PET/CT and USPIO-MRI in prostate cancer patients found that each modality identified potential LN metastases missed by the other.In particular, USPIO-MRI was superior at identifying suspicious LNs smaller than 5 mm. 17Unfortunately, an indisputable node-to-node comparison between in vivo imaging results and histopathology is hard to achieve, hampering the understanding of the identified differences.Because ePLND usually harvests more than 20 individual LNs, preparing the nodal material for histopathological examination by (1) finding all LNs within the resected adipose tissue specimens and (2) slicing all LNs into 3-mm-thick slices is very laborious and usually impossible to perform due to the high workload.As a result, small metastases are being missed. 18Using imaging as a guide to perform targeted sectioning of resection material could increase the detection rate and sensitivity of histopathology.
Furthermore, to comprehensively interpret the differences and agreements between the images from both modalities, a subnodal comparison is needed.With this aim, we conducted an explorative study comparing ex vivo 111 In-PSMA μSPECT images with high-resolution 7 T USPIO μMR images and histopathology of a selection of resected LN specimens from prostate cancer patients, allowing to assess the degree of correspondence.

Patient Population
The study population comprises patients from the Dutch DETECT trial (NCT4300673).Twenty men with histologically proven prostate cancer were included between September 2020 and April 2022.The study was approved by the local medical ethical review board (CMO Arnhem-Nijmegen), and written informed consent was obtained from all patients before enrolment.All patients had newly diagnosed prostate cancer and ≥1 PSMA-PET suspicious LN on preoperative PSMA-PET/CT.

Administration of Contrast Agents
Two days before surgery, patients received 2.6 mg/kg ferumoxtran-10 (Ferrotran; SPL Medical BV, Nijmegen, the Netherlands), through a slow intravenous drip infusion.The USPIO contrast agent ferumoxtran-10 (Ferrotran; SPL Medical BV, Nijmegen, the Netherlands) was used.This agent is available for clinical studies and in Named Patient Use Programs in the Netherlands and Switzerland.A large phase III international multicenter pivotal trial for European Medicines Agency approval is ongoing (EudraCT 2018-004310-18).The intravenous administration was performed over a period of 30-40 minutes and under medical supervision.One day before ePLND, patients received a mean single dose of 157 MBq (range, 151.8-164.2) 111 In-PSMA-I&T injected as an intravenous bolus and were observed for 1 hour postadministration.

Surgical Procedure and Tissue Preparation
Extended pelvic LN dissection was performed with radioguidance using a laparoscopic γ-probe (SOE-311-AL; Eurorad SA, Eckbolsheim, France) detecting radiation from 111 In-PSMA-I&T.Both in vivo and ex vivo (back table) probe measurements were taken to identify LNs with increased activity.Lymph nodes with a maximum ex vivo count of ≥10 counts per second (cps) were considered probe suspicious.Those LNs were considered LNs of interest (LNOIs).In addition, large palpable LNs without increased activity were also considered LNOIs.All LNOIs were marked with sutures and subject to elaborate additional imaging.Resected tissue was collected separately according to the anatomical resection site and fixated in formalin.After surgery, the specimens containing LNOIs were prepared for ex vivo μSPECT and μMR imaging.Specimens were removed from formalin, and digital photographs were taken in the intended orientation of ex vivo imaging.Specimens were carefully packed in cellophane and fixated on an MR-compatible Perspex sheet with a 200-μL Eppendorf tube filled with a solution of 111 In-PSMA-I&T as reference sample to match both MRI and SPECT images.Immediately after μSPECT and MRI acquisition, specimens were placed back in formalin to continue the fixation process.

Histopathological Evaluation
Histopathological evaluation was performed according to standard of care protocols with the addition of immunohistochemical PSMA staining.Lymph nodes of interest were separately numbered, sectioned, or sliced on different levels (5 mm), enclosed in tissue cassettes, and stained with hematoxylin-eosin and PSMA.

MR Image Evaluation and Coregistration With μSPECT
To provide iron-sensitive contrast for the interpretation of signal retention or attenuation within the lesions, T2*-fits of individual pixels of the multi-GRE images were used to calculate computed echo time (cTE) images 19,20 of 15 milliseconds.Ex vivo images were evaluated by the coordinating physician.Lymph nodes of interest (marked with suture) were matched to LNs on ex vivo MRI based on comparison with macroscopic photographs.They were recognized on the basis of both 3D MR image sets as a spheroid bordered structure on water-selective images without signal intensity on lipid selective GRE images.
The retainment of high MR signal within an LN on the 15 milliseconds image, either partially or completely, was classified as suspicious for containing metastases. 21,22Overlay of μSPECT reconstructed images on MR images was performed manually using the software assistant SAFIR technology (Fraunhofer MEVIS, Bremen, Germany).The Eppendorf tube functioned as a 3-dimensional structure of reference for image coregistration.

Image Analysis and Comparison With Histopathology
To substantiate the degree of correspondence between μSPECT and MR findings, a 4-level classification scheme was composed.This Likert scale is depicted in Figure 1 and describes the level of correspondence (LoC) with regard to the location and dimension of suspicious and nonsuspicious tumor tissue.The 4 categories range from poor correspondence (0%-25%; LoC 1), to moderate (25%-50%; LoC 2), to good (50%-75%; LoC 3), and excellent (75%-100%; LoC 4), with subcategories for appearance (a, b, c) (Fig. 1).Sizes of LNs were determined by measuring the longest axis in either 1 of 3 dimensions on lipid selective GRE images.The size of the suspicious part within an LN (high MR signal) was determined by measuring the longest axis in either 1 of 3 dimensions on the cTE 15 milliseconds water-excited GRE images.The assessment was performed by a trained researcher (M.S.) and an experienced radiologist (P.Z.).Any disagreement was resolved by discussion.
Histopathological data from all LNOIs were collected.Nodeto-node matching was achieved by manually matching macroscopically marked LNOI to nodes seen on ex vivo imaging and corresponding tissue cassette(s).Subnodal matching of histopathology and ex vivo imaging was difficult, so only histopathological presence or absence of metastatic tissue and PSMA expression was compared with imaging.All results are descriptive and depicted as numbers and percentages.

RESULTS
Twenty newly diagnosed prostate cancer patients were included.Baseline characteristics of these patients are summarized in Table 1.The median interval between USPIO administration and ePLND was 46 hours (interquartile range [IQR], 46-51).The median time between 111 In-PSMA-I&T administration and ex vivo μSPECT imaging was 28 hours (IQR, 26-31).No adverse events occurred during the administration of ferumoxtran-10 or 111 In-PSMA-I&T.

Evaluation of Ex Vivo μSPECT and USPIO-MRI
Based on suture localization, all 41 selected LNOIs could be successfully matched to an LN on ex vivo μMRI.The median size of these LNs was 18 mm (range, 5-55 mm).Of the 31 probe-suspicious LNOI, 27 (87%) showed increased activity at the site of the suture on μSPECT image analysis (Fig. 2).In the same 27 LNOIs, the evaluation of water-selective images (cTE = 15 milliseconds) showed a high MR signal intensity in (parts of ) the LNs compared with lipid tissue, making them suspicious of metastatic deposits.The median size of these deposits was 7 mm (range, 2-20).Histopathological evaluation confirmed the presence of PSMA-positive metastases in all 27 LNOIs (100%).Four LNOIs that were detected with the γ-probe did not indicate the presence of PSMA-avid metastases on μSPECT and on μMRI.Histopathological analysis of those lesions revealed no metastases.
None of the 10 large palpable, probe-negative LNOIs showed increased activity on μSPECT image analysis or suspicion on USPIO-μMRI.However, in 1 of the 10 large palpable LNOIs that were not suspicious on γ-probe measurements, histopathology revealed a PSMA-avid metastasis of 3 mm in a 42-mm LNs, which was neither identified with μSPECT imaging nor on μMRI, also not in retrospect.

Correspondence Between μSPECT and USPIO-MRI
Coregistration of μSPECT images with USPIO-enhanced water-selective multi-GRE images was successful for 41 LNOIs.Classification of LoC and characteristics of all LNOIs are shown in Table 2.The majority of lesions had an LoC 4 (37/41, 90%).Figure 3 shows an example of a partially affected metastatic LNOI (as based on ex vivo μMRI) with high correspondence of the SPECT signal in 3 dimensions (LoC 4b).In 5 of 41 LNOIs, coregistered images of USPIO-enhanced MRI and μSPECT showed suspicion of completely affected metastatic LNs, with no remaining healthy nodal tissue (LoC 4c) and median 90 maximum cps max (range, 59-599).An example of an LNOI with LoC 4c is displayed in Figure 4A.All LNOIs with LoC 4b and 4c were suspicious based on peroperative γ-probe measurements, and all contained PSMA-expressing metastases on final histopathology.A total of 14 LNOIs (35%) showed no suspicion of metastases on either μSPECT or ex vivo μMR imaging and were thus classified LoC 4a with a median maximum cps of 6 (range, 0-15 cps).The majority (10/14) of those LNs were marked based on palpable size.The remaining marked LNs (4/14) measured a maximum cps of 13, 13, 14, and 15 cps, respectively, and contained no metastases on histopathology.
A total of 3 LNOIs were classified as LoC 3 (7%) and 1 LNOI as LoC 2 (2%).Examples of both LoC 3 and LoC 2 are depicted in Figures 4B and 4C, respectively.All LoC 2 and LoC 3 lesions contained PSMA-expressing metastases on final histopathology.

DISCUSSION
Currently, there are 2 advanced noninvasive imaging techniques for LN staging of patients with prostate cancer: PSMA radionuclide imaging and USPIO-MRI, which outperform conventional imaging, that is, CT scanning or nonenhanced MRI. 23,24In this work, we evaluated the correspondence between these 2 modalities ex vivo on a subnodal level.It seemed to be technically feasible to perform high-quality, high-resolution imaging of resected LNs by both techniques.This allows to identify and relate LN anatomy and metastases on imaging with histopathology.
First, an important finding of this study is that both PSMA μSPECT and USPIO μMRI findings on a single-node level strongly correspond regarding the presence or absence of metastatic tissue on final histopathology.Only a 3-mm small partial metastasis was missed by both PSMA-μSPECT and μMRI.
Second, on subnodal level, meticulous coregistration of the μSPECT and μMR images demonstrated excellent correspondence regarding the presence of metastatic tissue and affected subnodal site in 90% (37/41) of the LNOI.Furthermore, both modalities strongly agreed in identifying both partially (LoC 4b) and completely (LoC 4c) affected LNs.We identified only 1 lesion with moderate (LoC 2) and no lesions with poor (LoC 1) correspondence.These findings not only verify concordance but also give insight in the subnodal appearance of both imaging technologies.As considerable uncertainties remain regarding the exact (patho)physiological mechanism of both USPIO contrast and PSMA tracers, exact explanations for the identified differences in LoC 2 and 3 lesions remain unknown. 25Although it is widely accepted that the lack of (or little) USPIO presence in metastatic LNs is due to the absence of USPIO-accumulating macrophages, the actual evidence regarding this hypothesis is limited. 12,13It still remains unclear which macrophage types accumulate USPIO.Meanwhile, the presence of partly non-PSMA-expressing tumor tissue (or other tissue, eg, fibrosis) might explain discrepancies. 14onetheless, our findings support the findings of Schilham et al, who found that the majority (73%) of PSMA-PET/CT suspicious LNs were also suspicious on USPIO-MRI. 14Interestingly, in their study, only 33% of the USPIO-MRI suspicious LNs were PSMA-PET positive.The small size of these LNs and the limitation of the spatial resolution of PET/CT are likely to contribute to this discrepancy, as LN metastases smaller than 5 mm are difficult to detect on PET/CT. 26This is also why true PET-positive nodal metastases (median, 4.0-13.6mm) are on average larger than false PET-negative metastases (median, 2.5-5.0 mm). 27he detection limit of USPIO-MRI is determined by the spatial resolution of the 3-dimensional MRI, which is 0.8-mm isotropic, resulting in the detection of many small suspicious nodes with short-axis diameter <3 mm LNs.For these small LNs, histopathological evaluation is difficult and time-consuming. 28In this work, we focused on a selection of LNs based on peroperative γ-probe measurements of excised tissue (31/41) and morphologically large LNs (10/41).Therefore, there was high a priori likelihood of an LN being positive on PSMA μSPECT and thus a selection bias in favor of SPECT-suspicious lesions.Second, although based on the same concept of PSMA expression, ex vivo γ-probe measurements are potentially more sensitive compared with in vivo PSMA-PET/CT, considering the smaller distance between the PSMA-expressing tissue and the γ-detector and the lack of background activity.
The concept of ex vivo imaging can support pathological evaluation of removed LNs, as determining the presence or absence of metastases in LN dissection specimens is a labor-intensive task.Extensive sectioning of the sample leads to far too much evaluation of the resected material, leading to the possible missing of small metastases and high costs.Ex vivo imaging results can provide quick and accurate information about the possible presence, amount, and location of suspicious lesions. 29The use of these images when cutting specimens can prevent small LNs and (micro)metastases within LNs. 18 This explorative study is not without limitations.Most importantly, due to its PSMA-PET-driven design, it comprises a selected subset of only PSMA γ-probe positive or large palpable LNs.This impedes a conclusion regarding the particularly interesting subgroup of small LNs that are missed in PSMA scanning and positive in USPIO-MRI.Second, although all lesions could be matched to histopathology on a node-to-node level, an exact subnodal anatomical comparison with histopathology was not possible.To better understand how both imaging modalities relate to pathophysiological processes in LNs, a systematic and meticulous correlation between imaging and pathology in larger series of LNs is needed. 22In such research, small LNs that are suspicious on in vivo USPIO-MRI and undetectable on PSMA-PET/CT are of particular interest.Third, our LoC classification is only semiquantitative, as SPECT imaging does not allow exact quantification of area of enhancement to compare to signal-enhanced area or volume on MRI.However, our 4-point Likert scale did allow a subnodal comparison between the imaging modalities.
In conclusion, this explorative, proof-of-concept study examined the correspondence between 2 advanced imaging modalities for nodal staging of prostate cancer on a subnodal level.Coregistration of both imaging modalities seems to be technically feasible, and both modalities showed excellent subnodal correspondence in the majority of lesions, in line with clinical histopathological readouts.

FIGURE 1 .
FIGURE 1. Four-point Likert scale for the level of correspondence between prostate-specific membrane antigen (PSMA) expression and ultrasmall superparamagnetic iron oxide (USPIO) particles in lymph nodes of interest (LNOIs).For PSMA μSPECT, the red color indicates the presence of PSMA-expressing tumor tissue.For USPIO μMRI, the black parts indicate loss of MR signal due to accumulation of USPIO contrast in the LNOI.The white parts indicate retaining of MR signal in tumor tissue.

FIGURE 2 .
FIGURE 2. Overview of all LNOIs.The flowchart demonstrates the numbers and percentages of LNOIs with respect to imaging results and histopathology on a single-node level.

Figure 1 ,
the drawings in the second column correspond to the LoC as described in Figure1; the red color indicates the presence of PSMA-expressing tumor tissue as seen on PMSA μSPECT.The black parts indicate loss of MR signal due to accumulation of USPIO contrast in the LNOI, and the white parts indicate retaining of MR signal in tumor tissue, as based on USPIO μMRI.†As measured on USPIO-enhanced ex vivo MRI.CPS, counts per second; IQR, interquartile range; LN, lymph node; LNM, lymph node metastasis; LNOI, lymph node of interest; LoC, level of correspondence; PSMA, prostate-specific membrane antigen; USPIO, ultrasmall superparamagnetic iron oxide particle.

FIGURE 3 .
FIGURE 3. Multiplanar reconstruction of an LNOI with excellent correspondence between PSMA expression and absence of USPIO particles.An LNOI depicted in 3 orthogonal orientations (pink [A, B, C], blue [D, E, F], and yellow [G, H, I]); one orientation per row in correspondence to the color-axes depicted in A, D, and G.The first column (A, D, and G) shows the lipid selective gradient echo (GRE) MR images (anatomic) where the border of the LNOI is indicated with the green dotted line.The second column (B, E, and H) shows the water-excited GRE MR images at a computed echo time of 15 milliseconds (iron sensitive).The third column (C, F, and I) shows the 15 milliseconds water-excited GRE images coregistered with PSMA μSPECT images.The round structure visible on the right corner of the last row is the Eppendorf tube for reference.

FIGURE 4 .
FIGURE 4. Typical examples of 3 individual LNOIs with different levels of correspondence, according to the 4-point Likert scale as described in Figure 1.Three examples of LNOIs with excellent (A), good (B), and moderate (C) correspondence between USPIO particle enhanced μMR and PSMA μSPECT.Left column shows the lipid selective GRE MR images (anatomic) where the border of the LNOI is indicated with the green dotted lines.The second column shows the water-excited GRE MR images at a computed echo time of 15 milliseconds (iron-sensitive).The third column shows the 15 milliseconds water-excited GRE images coregistered with PSMA μSPECT images.The fourth column shows 2 exemplary microscopic PSMA-stained slides of the LNOI.The LNOI in example A was classified as level of correspondence (LoC) 4c, example B as LoC 3a, and example C as LoC 2b in accordance to the description in Figure1.Of note, without proper alignment during workup, the orientation of histopathological slides was only visually matched to the imaging modalities.

TABLE 1 .
Patient Characteristics and Imaging Characteristics

TABLE 2 .
Characteristics of Lymph Nodes of Interest According to Level of Correspondence* *According to