A prospective study on inter-operator variability in semi-robotic software-based MRI/TRUS-fusion targeted prostate biopsies

Multiparametric magnetic resonance imaging (mpMRI) in combination with targeted biopsy (TB) has greatly improved the identification of clinically significant prostate cancer (csPCa) [1]. Thus, MRI/ultrasound-fusion biopsy (FB) has been widely introduced in the last decade. Software-based image-fusion has gained greatest acceptance of fusion techniques [2].

Although TB has increased the detection rate for csPCa, yet a considerable amount of csPCa still remains undetected by TB [3‐5]. This innovative approach of biopsy sampling represents a multi-step procedure involving different disciplines. Each step requires its own expertise which implies the occurrence of variations in the process. One of the preconditions for optimization of TB is the identification of its weaknesses to ensure no missing of csPCa.

The extent of inter-reader variability between radiologists and its implications for detection of prostate cancer (PCa) has already been described [6]. MRI result reporting from radiologists has also been shown to be of importance for biopsy performance [7]. Losses in accuracy in TB sampling have been analyzed in former research showing that experience of the urologist might also play an important role in cancer detection rates, CDRs [8, 9]. However, when evaluating inter-operator variability in FB, most studies compared CDR of urologists with different levels of experience on different patients [8‐10].

The aim of this prospective study was to assess the inter-operator variability and reproducibility of software-based fusion TB by targeting the same lesions by different urologists.

Characteristics of the study population are shown in Table 1. Overall, 155 patients received an MRI/TRUS-fusion biopsy and signed the informed consent for the study. Patients who either had no complete study biopsy by a second urologist (n = 11) or received a control-biopsy after focal therapy (n = 2) were excluded.

The overall CDR was 71.8% (102/142 patients). SB found 64.8% PCa compared to 62.0% found by both TB (TB1 + TB2) (p = 0.541). Total csPCa detection rate (csCDR) was 52.1% (41.5% SB vs. 43.0% TB, p = 0.851). Out of 102 cancer positive biopsies on a per-patient level, 11 PCa (10.8%) were detected exclusively by TB (TB1 + TB2) and 14 PCa only by SB (13.7%) (p = 0.690). Out of 74 patients with a csPCa (SB + TB), 13 patients (17.6%) had a csPCa finding solely in SB cores and 15 (20.3%) only in TB cores (TB1 + TB2) (p = 0.851). On a per-lesion level, TB detection rates were 55.8% (CDR) and 37.8% for csPCa.

There was no significant difference in CDR between TB1 and TB2 in all subgroups. The comparison of patient- and lesion-based TB1 and TB2 detection rates is shown in Online Resource 1. The lesion-based degree of agreement in detecting overall PCa (κ = 0.56) and in detecting nsPCa (κ = 0.56) between TB1 and TB2 was by definition “moderate”. Agreement in csPCa detection was by definition “substantial” (κ = 0.65) (Online Resource 2).

Figure 1 illustrates the number and types of mismatches between TB1 and TB2. In 20 MRI-suspicious lesions (11.6%), corresponding to 15 patients (10.6%), there was a mismatch between TB1 and TB2 in terms of overall cancer detection. Two out of 15 patients, whose lesions were only hit by one TB (TB1 or TB2), had a negative SB and no other positive lesions. In terms of csPCa detection, there was a patient-based mismatch of 14.8% (n = 21) between TB1 and TB2. Six out of those 21 mismatch patients (4.2% of total patients) had a benign or clinically insignificant finding (ISUP = 1) in the SB.

Of all PCa positive lesions, in which both, TB1 and TB2, were cancer positive (n = 76), upgrading to a csPCa, either in TB1 or TB2, occurred in 13.2% (n = 10).

The univariable and multivariable analyses of 112 lesions from the PCa positive patients for factors associated with mismatches revealed the size of the lesion (≤ 12 mm) described in the MRI as a predictive variable for mismatches (p < 0.001). Higher prostate volumes (> 41.21 ml) (OR 2.15, 95% CI 0.79–5.84, p = 0.184) did not significantly correlate with mismatches (Online Resource 3).

Analyses of the 102 PCa positive patients revealed that the size of the lesion was significantly smaller (p = 0.005) and prostate volume was significantly bigger (p = 0.014) in the group of patients who were only cancer positive in SB (n = 14).

Adverse events occurred in 7.0% (n = 10) of patients. Hematuria was detected in 2.0% (n = 3), urinary retention in 2.0% (n = 3), rectal bleeding in 1.4% (n = 2) and fever in 1.4% (n = 2).

Despite the superiority of FB compared to SB, many high-risk PCa still remains undetected by TB as shown by Ahdoot et al. [3]. They showed a misclassification rate of up to 13.6% of all csPCa bearing patients if SB would have been omitted [3]. In our study cohort, 17.6% of PCa patients with ISUP-score ≥ 2 were misclassified by TB. Many factors have already been investigated which might influence FB [6‐10]. These findings leave us the consideration of why the biopsy result of a suspicious lesion is negative.

The methodical approach of investigating inter-operator variability in the same patient by two different urologists to eliminate all confounding factors of procedure comparison has so far not been undertaken.

A key finding of our present study is that a considerable number of mismatches in PCa detection as well as in csPCa detection between both urologists could be observed. Although overall PCa and csPCa detection rate of TB1 and TB2 did not differ significantly on a per-patient and on a per-lesion level, a discrepancy in csPCa finding occurred in 14.8% of the study population. In total, a csPCa finding could have been missed in up to 4.2% of all patients, if the TB had been carried out only by one of the urologists. The remaining csPCa mismatch patients would have been covered by the SB, which accounts for 10.6%, underlining the importance SB still has in this setting. This result also suggests that more than two cores should be taken from each target lesion in FB. Even though the number of TB cores was not a predictor for occurrence of mismatches here, several trials showed that up to 10% of patients would benefit from more than two cores per lesion [11].

Compared to Ahdoot et al., we identified a similar yield of PCa (62.0% vs. 51.5%) and csPCa (43.0% vs. 37.8%) with TB in our study [3]. Although detection rates between both TB did not differ significantly, the lesion-based level of agreement was not optimal. In total, 20 lesion-based PCa detection discrepancies between TB1 and TB2 were found in our study. Half of those 20 lesions (n = 10) comprised a clinically significant cancer finding, emphasizing the imperfect reproducibility of TB even with the remarkable assistance of a biopsy platform.

In search of potential factors influencing the occurrence of mismatches between urologists, a smaller size of the MRI-lesion was revealed as a significant predictor (OR = 9.19). The size of the lesion is also associated with the likelihood of both urologists missing the target, which supports the assumption that smaller PCa lesions are less likely to be identified by TB. This finding is in agreement with the recent study of Baco et al. describing a reduced csPCa detection rate of 50% for lesions < 0.5 ml vs. 76% for lesions > 1 ml in size [12, 13]. It has been suggested that a perfect fusion of both images is necessary to reliably hit smaller targets. A small error in prostate and lesion boundary segmentation can already have a major impact on successful targeting [13]. The fact that both urologists in our study needed to carry out the delineation of the prostate in TRUS images, a procedure which requires high precision and is thus a source of targeting error, could explain these findings. The number of mismatches might be even higher if contouring of prostate and lesions in MRI-images was also done by each urologist separately. As discussed by Tay et al., ultrasound-segmentation of the prostate necessitates a smooth and even sweep by the probe to avoid any displacements or rotation which may affect the shape of the 3D-construct, thus avoiding inaccurately displaying the target lesion. Sudden movements of the patients during the sweep, for example due to discomfort as well as prostate deformation by the application of too much pressure with the probe, can also alter the shape of the 3D-construct [13]. Although fusion platforms attempt to correct these alterations by using elastic registration algorithms and motion compensation, our results suggest the need for further optimization in this field [14]. Of particular importance, a deviation of the needle from the intended and predefined core path, for example due to its asymmetric bevel, is less likely to be compensated in smaller lesion sizes [13]. However, clinicians are partly able to adapt to this veering effect as they become more experienced over time and a (semi-)robotic needle guidance might further reduce the user-dependent effect. In contrast to previously published similar studies, we did not observe a large impact of the urologist’s experience on cancer detection [8‐10].

No significant correlation was shown between prostate volume and the occurrence of mismatches, which might be due to the rather small sample size of our cohort. However, inverse association of prostate volume with PCa detection by FB was previously described [15, 16]. It is postulated that an increased prostate volume is associated with the deformation of the prostate during biopsy procedure leading to registration errors. Furthermore, the increased depth of the target lesion as may be found in an enlarged prostate is likely to be associated with increase in deviation of biopsy path [17].

The number of adverse events during and after the procedure are comparable to those in other studies [18].

Interpreting our results, a key limitation is the possibility of different surgical conditions for the first and second surgeon. It is suggested that the accuracy of hitting the target lesion on the real-time TRUS image during the second TB is decreased by tissue swelling caused by the first biopsy procedure. Regarding the discrepancies in csPCa finding, considerations to heterogeneity of tumor lesions should be made. Aihara et al. revealed in PCa specimens that with increased lesion size multiple grades of PCa can be present which are arranged in heterogeneous and unpredictable geographic interrelationships [19]. Therefore, evaluation of each surgeon’s accuracy based on the grade of PCa might be limited. Despite the large number of different urologists taking biopsies, our results are still valuable since they reflect the real-world practice.

This study demonstrates that reproducibility and precision of targeting lesions suspicious for PCa is still limited, even with the high-level support of a semi-robotically software-based fusion biopsy platform. Although the detection rate of PCa and csPCa can be markedly improved by FB, discrepancies in biopsy results between individual urologists can still be observed. This insight should serve as an incentive for further improvements in image fusion, segmentation, needle guidance as well as automatization of the procedure so that even unexperienced clinicians are able to reliably hit a small lesion in an enlarged prostate.