Selecting patients for magnetic resonance imaging cognitive versus ultrasound fusion biopsy of the prostate: A within‐patient comparison

Abstract Objectives To compare overall agreement between magnetic resonance imaging (MRI)–ultrasound (US) fusion biopsy (FB) and MRI cognitive fusion biopsy (CB) of the prostate and determine which factors affect agreement for prostate cancer (PCa) who underwent both modalities in a prospective within‐patient protocol. Patients and Methods From August 2017 to January 2021, patients with at least one Prostate Imaging Reporting & Data System (PI‐RADS) 3 or higher lesion on multiparametric MRI underwent transrectal FB and CB in a prospective within‐patient protocol. CB was performed for each region of interest (ROI), followed by FB, followed by standard 12 core biopsy. Patients who were not on active surveillance were analysed. The primary endpoint was agreement for any PCa detection. McNemar's test and kappa statistic were used to analyse agreement. Chi‐square test, Fisher's exact test and Wilcoxon rank sum test were used to analyse disagreement across clinical and MRI spatial variables. A multivariable generalized mixed‐effect model was used to compare the interaction between select variables and fusion modality. Statistics were performed using SAS and R. Results Ninety patients and 98 lesions were included in the analysis. There was moderate agreement between FB and CB (k = 0.715). McNemar's test was insignificant (p = 0.285). Anterior location was the only variable associated with a significant variation in agreement, which was 70% for anterior lesions versus 89.7% for non‐anterior lesions (p = 0.035). Discordance did not vary significantly across other variables. In a mixed‐effect model, the interaction between anterior location and use of FB was insignificant (p = 0.411). Conclusion In a within‐patient protocol of patients not on active surveillance, FB and CB performed similarly for PCa detection and with moderate agreement. Anterior location was associated with significantly higher disagreement, whereas other patient and lesion characteristics were not. Additional studies are needed to determine optimal biopsy technique for sampling anterior ROI.


| Diagnostic imaging
The majority of mpMRIs were performed at 3-Tesla with an endorectal coil using the technical standards set forth by PI-RADS v2. All mpMRI, including the few not performed at our institution, were interpreted by two diagnostic radiologists (BF, FC) with extensive mpMRI experience using PI-RADS v2 criteria. Prostate and ROI contouring was performed using Dynacad software (Invivo, Philips, Best, Netherlands).

| Biopsy procedure details
All biopsies were performed by three fellowship-trained urologic oncologists (JJL, SI, CLA) with extensive experience in fusion and standard TRUS biopsy. One of the surgeons performed FB prior to this study in training, while the other two surgeons had been performing FB for 2 years prior to initiation of the study. Informed consent was obtained, and biopsies were conducted per protocol, not as standard of care. The protocol is as follows. First, a TRUS was performed (BK Medical, Peabody, MA), and two cores from the targeted ROI on MRI were obtained using the CB technique, sampling either a TRUSvisible hypoechoic ROI corresponding to the region identified by MRI or the general area specified by MRI if no hypoechoic ROI could be identified. Second, the MRI target ROI and prostate contour were coregistered with the TRUS image using the UroNav FB device (InVivo, Philips, Best, Netherlands), and two cores were obtained from the ROI. If multiple ROIs were present, CB of all ROIs was performed first.
FB followed in the standard fashion. Finally, a 12-core standard sextant systematic biopsy was obtained.

| Outcomes
Overall, cancer detection was the primary outcome and defined as any Gleason 3 + 3 or higher found in one or more cores for the ROI on biopsy final pathology. High-grade cancer detection was a secondary outcome and defined as any Gleason 3 + 4 or higher found in one or more cores for the ROI on biopsy final pathology.

| Statistical analysis
McNemar's test was used to analyse overall agreement of any PCa detection between CB and SB.
Agreement was then assessed across specific pre-biopsy clinical and MRI spatial parameters. These included clinical variables that are known or suspected to affect cancer detection rate, namely, age, ethnicity, family history of PCa, digital rectal exam (DRE) and previous biopsy. We also included MRI characteristics including PI-RADS score, prostate volume, ROI maximum dimension, transverse dimension, anteroposterior dimension, craniocaudal dimension, and volume, and ROI location (base, apical and anterior). Correlation between agreement for any cancer detection and each variable was tested using chisquare test, Fisher's test or Wilcoxon rank sum test where appropriate. A p value of <0.05 was considered significant.
Variables that significantly affected agreement were then selected for a separate secondary multivariable analysis to compare cancer detection rate between fusion modalities. A generalized mixed-effect model was used. To compare CB and FB directly, all ROIs were counted twice within the same model. Individual ROIs had a unique identifier, and this was included as a random effect variable.
Fusion modality used (CB or FB) was included as a fixed effect. Interactions between significant variables from the above subgroup analysis and fusion modality were tested. R (version 3.63) and SAS version 9.4 were used.

| Patient and tumour characteristics
Ninety patients and 98 ROIs were included in the study. The median age was 68, and median PSA was 7 ng/ml. Sixty-four percent of patients were diagnosed with any cancer, and 52% of patients were diagnosed with Gleason grade group 2 or higher. High-grade cancer detection rate for both targeted biopsy techniques amongst all ROIs was 48%. High-grade cancer detection rate amongst all ROIs was 40% for FB, and 43% for CB. Tables 1A and 1B summarize patient-level and ROI-level characteristics. Fifty-three percent of all patients had a prior negative biopsy.

| Concordance
Overall, concordance between FB and CB was 85.7% for all ROIs. FB detected cancer where CB did not in nine ROIs (9%). CB detected cancer where FB did not in five ROIs (5%) (

| Multivariable analysis with interaction variables
As anterior location was the only significant variable from the agreement analysis, this was selected and included in the generalized mixedeffect model as part of an interaction term between anterior location and fusion modality. Neither the modality main effect nor the interaction terms were significant, suggesting neither modality was superior for the detection of PCa in our cohort. As expected, anterior location was statistically significant for PCa detection (p = 0.015). The odds ratio of any cancer detection in anterior lesions compared with non-anterior lesions were 3.06 (95% confidence interval [CI] 1.3, 7.4) ( Table 3).

| DISCUSSION
This study found that in a within-patient protocol, there was moderate agreement for the detection of any PCa between CB and FB all patients One possibility as to why disagreement is higher amongst anterior lesions specifically is that there is more deformation of tissue to reach these targets due to increased distance from needle entry point. The advent of targeted MRI visible lesions has likely not surmounted the challenge of sampling the anterior zone of the prostate with transrectal biopsy approaches. This is reinforced by the higher rate of PCa detection in anterior lesions on multivariable analysis, as patients may have had anterior cancers that were missed with prior negative standard template biopsy as is known to occur commonly. 14,15 For this reason, urologists must continue to give special attention to patients with MRI visible anterior tumours to avoid diagnostic sampling error. Further studies are needed to determine if these patients may be better served by alternate biopsy techniques, such as MRI-guided in-bore or transperineal biopsy, which are becoming increasingly utilized at tertiary referral centers. [16][17][18][19] As we have seen in this study, FB does not appear to be superior to CB in this setting, but the high rate of disagreement suggests that negative results of any transrectal FB should be interpreted with caution. amongst anterior ROIs might decrease with increased sampling. More