The Potential of Iron Oxide Nanoparticle-Enhanced MRI at 7 T Compared With 3 T for Detecting Small Suspicious Lymph Nodes in Patients With Prostate Cancer

Background Accurate detection of lymph node (LN) metastases in prostate cancer (PCa) is a challenging but crucial step for disease staging. Ultrasmall superparamagnetic iron oxide (USPIO)–enhanced magnetic resonance imaging (MRI) enables distinction between healthy LNs and nodes suspicious for harboring metastases. When combined with MRI at an ultra-high magnetic field, an unprecedented spatial resolution can be exploited to visualize these LNs. Purpose The aim of this study was to explore USPIO-enhanced MRI at 7 T in comparison to 3 T for the detection of small suspicious LNs in the same cohort of patients with PCa. Materials and Methods Twenty PCa patients with high-risk primary or recurrent disease were referred to our hospital for an investigational USPIO-enhanced 3 T MRI examination with ferumoxtran-10. With consent, they underwent a 7 T MRI on the same day. Three-dimensional anatomical and T2*-weighted images of both examinations were evaluated blinded, with an interval, by 2 readers who annotated LNs suspicious for metastases. Number, size, and level of suspicion (LoS) of LNs were paired within patients and compared between field strengths. Results At 7 T, both readers annotated significantly more LNs compared with 3 T (474 and 284 vs 344 and 162), with 116 suspicious LNs on 7 T (range, 1–34 per patient) and 79 suspicious LNs on 3 T (range, 1–14 per patient) in 17 patients. For suspicious LNs, the median short axis diameter was 2.6 mm on 7 T (1.3–9.5 mm) and 2.8 mm for 3 T (1.7–10.4 mm, P = 0.05), with large overlap in short axis of annotated LNs between LoS groups. At 7 T, significantly more suspicious LNs had a short axis <2.5 mm compared with 3 T (44% vs 27%). Magnetic resonance imaging at 7 T provided better image quality and structure delineation and a higher LoS score for suspicious nodes. Conclusions In the same cohort of patients with PCa, more and more small LNs were detected on 7 T USPIO-enhanced MRI compared with 3 T MRI. Suspicious LNs are generally very small, and increased nodal size was not a good indication of suspicion for the presence of metastases. The high spatial resolution of USPIO-enhanced MRI at 7 T improves structure delineation and the visibility of very small suspicious LNs, potentially expanding the in vivo detection limits of pelvic LN metastases in PCa patients.

Initial USPIO-enhanced MRI studies in PCa were performed at 1.5 T. 3,16 Magnetic resonance imaging at a higher (3 T) field strength improved the image quality of the technique. 17In addition, 3D acquisition techniques at 3 T enabled imaging at high isotropic image resolutions (0.85 Â 0.85 Â 0.85 mm), which was used at the time of the reintroduction of USPIO-enhanced MRI. 18In a first direct comparison study between USPIO-enhanced MRI at 3 T and PSMA-PET/CT in the same patients, MRI detected more and smaller suspicious LNs than PSMA-PET/CT did. 19he challenging next step in body MRI is moving to an ultra-high magnetic field strength of 7 T. Exploiting its higher sensitivity by acquiring an even higher spatial resolution than reachable at 3 T, USPIO-enhanced MRI at 7 T could allow improved detection of LNM, which may alter disease management, both in primary PCa and in recurrent disease. 20,21][26] To assess the potential of improved detection of suspicious LNs with USPIO-enhanced MRI at 7 T over 3 T, we compared 7 T USPIO-enhanced MRI with 3 T USPIO-enhanced MRI in a cohort of 20 patients with PCa.

Participants
This single-center, nonrandomized, prospective trial was approved by the institutional review board (no.16 7214 BO), and written informed consent was obtained from all participants.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 EMA approval is ongoing (EudraCT 2018-004310-18).For this study, patients with primary or recurrent PCa, who were referred to our medical center for an investigational USPIO-enhanced 3 T MRI examination due to deemed high risk of LNM (by primary risk classification 20 or PSA level after therapy 21 ), were included between March 2014 and February 2015.The patients provided written informed consent to undergo a 7 T USPIO-enhanced MRI on the same day as well.

USPIO-Enhanced MRI at 3 T and 7 T
All participants received an intravenous administration of ferumoxtran-10 (dosage 2.6 mg/kg) 24-36 hours before 3 T and 7 T MRI.High-resolution, 3-dimensional anatomical, and T2*-weighted images were acquired at 3 T (MAGNETOM Prisma-fit; Siemens Healthineers, Erlangen, Germany) and within 5 hours on the same day at 7 T (MAGNETOM 7 T; Siemens Healthineers, Erlangen, Germany).Negligible difference in nodal contrast accumulation was to be expected for these 2 scans within this time frame. 27The scanned area included at least the area from the aortic bifurcation to the bladder neck (see Table 1 for scanning parameters).

Image Evaluation and Annotation
After anonymization, all scans were independently evaluated by 2 experienced abdominal radiologists (A.S.F. with 6.5 years and T.H. with 3 years, R1 and R2, respectively) who were blinded for clinical parameters.First, the 3 T studies were presented to the readers.To avoid recollection, the 7 T scans were assessed in a blinded random order at least 3 months later.R1 is an abdominal radiologist with initial experience for reading USPIO-enhanced imaging; R2 is an expert in assessing clinical USPIO-enhanced MRI.Both readers were asked to annotate the LNs suspicious for harboring nodal metastases on the basis of both the anatomical and the T2*-weighted 3D datasets with freedom to scroll in 3 dimensions and with optional maximum intensity projection.The water-selective iron-sensitive T2*-weighted images were used to identify suspicious LNs (3 T: multiecho data image combination, 7 T: computed echo time images from multigradient echo); LNs with a homogeneous residual high signal intensity or those with partial high signal components were deemed suspicious for harboring metastases.The reconstructed water and lipid images (3 T: volumetric interpolated breath-hold examination Dixon, 7 T: gradient echo) were used for anatomical correlation.The nodes were annotated on T2*-weighted MRI and scored on a 5-point level of suspicion (LoS) scale (metastases very unlikely [1], unlikely [2], equivocal [3], likely [4], and very likely [5]).The readers had the assignment to find all potentially suspicious LNs (defined as LoS ≥3).Healthy LNs did not have to be scored, although the scores of LoS 1 and 2 could be used to indicate an LN or other small spherical structure retaining signal on T2*-weighted MRI to deem it benign (eg, a slightly gray LN or a ganglion).
Lymph node annotations from both readers were imported independently into an adapted dedicated viewer (Mevislab; Mevis Medical Solutions, Bremen, Germany).The annotations of both readers could be turned on and off and projected over the patient's images, enabling precise matching between the annotations of both readers.When both readers independently annotated an LN and scored it with LoS 3-5, it was considered "suspicious."When both readers independently detected an LN with a score of LoS 4-5, it was considered "highly suspicious."An independent reader (A.V.) performed matching of annotations and nodal The subset R1 and R2 contains the LNs that were annotated by both readers; the total number encompasses all LNs annotated by either of the 2 readers.LoS, level of suspicion.size measurements in 3 orthogonal directions (noted as short axis diameter and ellipsoid volume).R1 assessed the image quality of all 3 T and 7 T images (see Supplementary Material, http://links.lww.com/RLI/A888).

Statistical Analysis
To compare the detection of suspicious LNs on 3 T and 7 T MRI, the number of suspicious LNs was analyzed per participant by Wilcoxon matched pairs signed rank tests.Mann-Whitney U tests were performed to compare the size of annotated LNs.The LN size is compared between the LoS score groups by Dunn multiple comparison tests after Kruskal-Wallis nonparametric analysis of variance.A subgroup of LNs that were scored on both field strengths by both readers was analyzed on LoS scores with Wilcoxon matched pairs signed rank tests.A P value <0.05 indicated a statistically significant difference.

Participants
Twenty patients were included with a mean age of 63 years (range, 50-74 years).Participants with primary PCa (n = 7) and patients with recurrent disease (n = 13) had mean serum prostate-specific antigen levels of 33.1 ng/mL and 2.8 ng/mL, respectively (Table 2).In all 20 participants, the infusion of the contrast agent was without any adverse events, and they underwent both MRI examinations successfully.The overall image quality and the delineation of anatomical structures on the T2*-weighted images were scored significantly higher on 7 T, with a median score for overall image quality of 4 and 5 for 3 T and 7 T, respectively (P < 0.001) (Fig. 1).For more extensive elaboration of analysis of image quality and a movie comparing 3 Twith 7 T data, see the Supplementary Materials, http://links.lww.com/RLI/A889.

Number of LNs
The total number of annotated LNs by either of the 2 readers was 599 at 7 T and 408 at 3 T (Table 3).The total number of LNs observed per patient (across all levels of suspicion) was higher at 7 T than at 3 T, with a median number (and range) of LNs annotated by at least 1 reader of 23.5 (2-88) versus 17.5 (2-45), P < 0.001 (Fig. 2A).Subgroup analysis showed significant differences in median number of LNs per patient for both primary (7 T: 20 vs 3 T: 17, P = 0.03) and recurrent disease (7 T: 28 vs 3 T: 18, P = 0.009), respectively.

LN Size
Lymph nodes annotated by at least 1 reader (across all levels of suspicion) had a median short axis at 7 T and 3 T of 2.5 and 2.6 mm  (Fig. 3), respectively (not significantly different, P = 0.78).Subgroup analysis showed no significant differences in median short axis of LNs for primary (7 T: 2.5 vs 3 T: 2.7 mm, P = 0.24) or recurrent disease (7 T: 2.5 vs 3 T: 2.5 mm, P = 0.61), respectively.
Figure 5 illustrates the size and LoS of all annotated LNs.The short axis diameter shows great overlap between the LoS groups.At 7 T, the median value for LoS 5 shows a significant difference to both LoS 4 (P = 0.002) and LoS 3 (P = 0.02).At 3 T, the median value for LoS 5 shows a significant difference to LoS 3 (P < 0.001).

Interreader Agreement
R1 and R2 both annotated more LNs on 7 T MRI (474 and 284) compared with 3 T MRI (344 and 162, respectively), see Table 3.Both readers had 98 matching LNs of a total of 408 nodes at 3 T and 159 matching LNs of 599 at 7 T.The individual LoS scores for LNs for both readers can be found in Table 4. R1 gave higher LoS scores for both field strengths (P < 0.001 for 7 T and 3 T).

Matched LNs Between Field Strengths
Lymph nodes that were annotated by both readers were matched between field strengths (example in Fig. 1).From the 79 suspicious LNs found on 3 T images, almost all were also annotated on the 7 T images: either by both readers (51) or one of them (21).Of 116 suspicious LNs on 7 T images, 33 were not scored on 3 T images by either of the 2 readers (median short axis of 2.2 mm; range, 1.5-3.0mm).
Fifty-eight LNs were scored by both readers on both field strengths and therefore further analyzed.There was no difference in size for the matched LNs on both field strengths (short axis mean: 3.64 [7 T], 3.40 [3 T]; P = 0.11).The interpretation of one reader did differ between field strengths in these matched LNs, as R2 gave significantly higher LoS scores for the same LNs on 7 T images compared with 3 T (P < 0.001, R1: P = 0.60).

DISCUSSION
This is the first study in which 7 T USPIO-enhanced MRI is compared with 3 T USPIO-enhanced MRI to detect LNs suspicious of metastases in the same patients with PCa.At 7 T, high-quality images allow the detection of more and especially more small (<2.5 mm) suspicious LNs.More highly suspicious LNs (LoS 4/5) were found per patient on 7 T compared with 3 T (P = 0.02).We showed that suspicious nodes in PCa are generally very small, with 44% of the suspicious LNs found on 7 T measuring <2.5 mm in short axis.
With different radiofrequency coils and pulse sequence possibilities at both field strengths, pulse sequences and sequence parameters were not the same for this comparison between 7 T and 3 T, but optimized to imaging on either 7 T or 3 T to exploit the advantages of that specific field strength.Examinations at 7 Twith its substantially higher spatial resolution (voxel size being a factor 2.14 higher than 3 T: 0.29 mm 3 vs 0.61 mm 3 ) enabled us to detect more LNs with a short axis diameter below 2 mm.Although the short axis diameters of suspicious nodes did not differ significantly between 7 and 3 T, incorporating 3 spatial dimensions by measuring nodal volumes did reveal a significant difference in size between suspicious nodes detected at 7 T versus 3 T. Matching annotated LNs between field strengths showed that many of these very small LNs were not annotated on 3 T, possibly just due to their small size or due to being closely adjacent to and thereby  indistinguishable from neighboring structures.As matched LNs did not show a difference in measured short axis diameter between field strengths, this indicates that there is no consistent measurement error due to the image quality or resolution.
The high number of suspicious LNs per patient is not unexpected, as the pelvis is known to hold many LNs (range, 19-91 LNs per healthy volunteer on 7 T MRI). 24The large variability in the number of annotated LNs per patient in this work represents the actual differences in patients at high risk for metastatic disease: 3 participants had more than 10 suspicious LNs, 3 patients had none, and in the remaining participants, the number of suspicious LNs ranged from 1 to 7. The number of annotated LNs located (partly) outside the standard ePLND resection field is in line with previous work 19 and underlines the diagnostic and therapeutic challenges of ePLND.
As a large proportion of suspicious LNs in PCa were very small, the ability to identify these LNs could have important clinical implications.As a comparison, in breast cancer, direct ultrasound-guided fine-needle aspiration or core biopsy is recommended in axillary LNs with a cutoff point of maximum cortex thickness as small as 2.3 mm to achieve a high sensitivity of 95% and accepting a limited specificity of 44%. 28A positive result directly changes further diagnostics. 29In our study, there was a large overlap in short axis of annotated LNs between LoS groups.The idea that large LNs are suspicious and small LNs are not suspicious might no longer be valid in PCa, as most suspicious LNs were very small.The benefit for the patient in detecting more and more small suspicious LNs will most probably not be found in extending pelvic LN dissections, as its current impact on overall survival is already under debate, 1,2 and removal of all small suspicious nodes will be very challenging.1][32] Although PSMA-PET/CT is of increasing clinical benefit for image-guided therapy and has the advantage of simultaneous assessment of distant (bone) metastases, its spatial resolution for LN detection is limited compared with the spatial resolution of 7 T USPIO-enhanced MRI.
R1 annotated many more LNs on both 7 T and 3 T examinations compared with R2, possibly because of the difference in experience with USPIO-enhanced MRI in combination with a continuous scale of nodal signal intensity.Annotating and scoring LNs based on the presence or absence of USPIO contrast-loaded macrophages enable a functional assessment that is no longer based on size and morphological characteristics.This requires, however, a different approach to "reading" images.The interpretation of USPIO accumulation with its different shades of gray without quantifying signal intensity requires a learning process.Of note is the increased confidence of the experienced USPIO-MRI reader R2 in scoring field strength-matched suspicious nodes, presumably because of the improved image quality at 7 T. Recently, Driessen et al 33,34 have described a new paradigm for evaluating USPIO-enhanced MRI of the head and neck region, incorporating differences in retained signal intensity of nodes.Aided by an extensive workflow for node-to-node matching of in vivo detected LNs with histopathology of resected nodes, 34 a new reading algorithm was designed incorporating the signal intensity of LNs compared with the surrounding fat, improving discrimination between metastatic and nonmetastatic LNs. 33,34This algorithm potentially also improves the agreement between readers in future studies if lipid tissue is not fully suppressed (not excited) in the multiecho gradient echo pulse sequence.
An important limitation of this study is that there is no histopathological reference standard for the suspicious LNs.Four of the 20 participants underwent ePLND after USPIO-enhanced imaging, but without a dedicated workflow for node-to-node matching of surgical specimens to pathological review of nodes. 34,26On a patient level, the USPIO-enhanced MRI results corresponded to the histopathological outcome: 3 patients with suspicious LNs on 3 T and 7 T MRI had LNMs on histopathology; the other patient who had no suspicious LNs on 7 T imaging and 1 suspicious LN on 3 T had no LNMs on histopathology.In absence of matched histopathology, we have chosen to deem LNs suspicious in case both readers annotated the LN with a high LoS.The sensitivity and specificity of 7 T USPIO-enhanced MRI to detect especially small LNMs in PCa remain to be investigated, this study however gives an indication of what to expect of high field strength MRI.
In conclusion, this first comparison of 7 T USPIO-enhanced MRI with 3 T USPIO-enhanced MRI in the same cohort of PCa patients reveals an increased detection of small LNs suspicious for harboring metastases on 7 T MRI.The high resolution of USPIO-enhanced 7 T MRI potentially expands the in vivo detection limits of pelvic LNM in PCa patients and can offer insight in pelvic LNM distributions.This may provide a new gateway to personalized image-guided therapy.

FIGURE 1 .
FIGURE 1. Example of USPIO-enhanced MRI of a patient with advanced prostate cancer at 7 T and 3 T in transversal planes.A big LN (short axis 7.5 mm, arrows) was scored by both readers with an LoS of 5 on both 7 T and 3 T.The smaller LN (short axis 3.0 mm, arrowheads) was scored suspicious with LoS 5 by both readers on 7 T. On 3 T, possibly due to its size or appearance, this node was not annotated by either of the readers.

FIGURE 3 .
FIGURE 3. Size distribution of annotated LNs per field strength (all levels of suspicion).Bars are cumulative, showing the LNs with their short axis, split by reader.

FIGURE 2 .
FIGURE 2. A, All annotated LNs per patient for 7 T and 3 T, with all patients (n = 20) and divided into primary (n = 7) and recurrent PCa (n = 13).B, Suspicious LNs (both readers LoS ≥3), for 7 T and 3 T. p indicates primary prostate cancer; r, recurrent disease.

FIGURE 5 .
FIGURE 5. Annotated LNs and their size ordered by LoS, with median value per LoS score group indicated.Scores of 1 and 2 were only given to mark a notable but unsuspicious LN, not incorporating any nodes without signal intensity on T2*-weighted imaging.*P < 0.05, **P < 0.005, ***P < 0.001.

FIGURE 4 .
FIGURE 4. Size distributions of suspicious LNs at 7 T and 3 T, noted as percentage of the total number (116 at 7 T and 79 at 3 T), ordered by short axis (A) and volume (B). 3 T data (semitransparent) is in front of 7 T data, with overlapping bars in purple.

TABLE 2 .
Demographic Data of Participants

TABLE 1 .
Overview of Pulse Sequence Parameters for 7 T and 3 T MR Acquisitions

TABLE 3 .
Number of Annotated LNs on Both Field Strengths, Given for Each Reader Separately and Divided Further on Level of Suspicion

TABLE 4 .
Confusion Matrices of Interreader Agreement on 159 From 599 Annotated Nodes at 7 T and on 98 From 408 Annotated Nodes at 3 T Scores of 1 and 2 were only given to mark a notable but unsuspicious LN.LNs with scores ≥ Los3, in bold, were considered suspicious.