Trans-rectal ultrasound visibility of prostate lesions identified by magnetic resonance imaging increases accuracy of image-fusion targeted biopsies

Abstract

Purpose

To compare the diagnostic yield of targeted prostate biopsy using image-fusion of multi-parametric magnetic resonance (mp-MR) with real-time trans-rectal ultrasound (TRUS) for clinically significant lesions that are suspicious only on mp-MR versus lesions that are suspicious on both mp-MR and TRUS.

Methods

Pre-biopsy MRI and TRUS were each scaled on a 3-point score: highly suspicious, likely, and unlikely for clinically significant cancer (sPCa). Using an MR-TRUS elastic image-fusion system (Koelis), a 127 consecutive patients with a suspicious clinically significant index lesion on pre-biopsy mp-MR underwent systematic biopsies and MR/US-fusion targeted biopsies (01/2010–09/2013). Biopsy histological outcomes were retrospectively compared with MR suspicion level and TRUS-visibility of the MR-suspicious lesion. sPCa was defined as biopsy Gleason score ≥7 and/or maximum cancer core length ≥5 mm.

Results

Targeted biopsies outperformed systematic biopsies in overall cancer detection rate (61 vs. 41 %; p = 0.007), sPCa detection rate (43 vs. 23 %; p = 0.0013), cancer core length (7.5 vs. 3.9 mm; p = 0.0002), and cancer rate per core (56 vs. 12 %; p < 0.0001) = 0.0001), respectively. Highly suspicious lesions on mp-MR correlated with higher positive biopsy rate (p < 0.0001), higher Gleason score (p = 0.018), and greater cancer core length (p < 0.0001). Highly suspicious lesions on TRUS in corresponding to MR-suspicious lesion had a higher biopsy yield (p < 0.0001) and higher sPCa detection rate (p < 0.0001). Since majority of MR-suspicious lesions were also suspicious on TRUS, TRUS-visibility allowed selection of the specific MR-visible lesion which should be targeted from among the multiple TRUS suspicious lesions in each prostate.

Conclusions

MR-TRUS fusion-image-guided biopsies outperformed systematic biopsies. TRUS-visibility of a MR-suspicious lesion facilitates image-guided biopsies, resulting in higher detection of significant cancer.

Introduction

Traditional trans-rectal ultrasound (TRUS)-guided systematic random biopsies may under-diagnose clinically significant cancer and over-detect clinically insignificant cancer [1–3]. Diagnostic accuracy of TRUS depends upon the operator’s knowledge of, and expertise with, ultrasound technology, since many benign intra-prostatic lesions can appear hypo-echoic, which therefore is not a specific sign of cancer [1, 3, 4].

Recent evidence suggests that multi-parametric MR (mp-MR) may better visualize clinically significant cancer, and mp-MR, with or without image-fusion with real-time TRUS, is being evaluated for targeted biopsy guidance [5–10]. Emerging data indicate that, depending on the clarity of MR findings, MR-TRUS image-fusion targeted biopsies may better diagnose higher grade and/or higher volume cancer compared to systematic random biopsies, while utilizing fewer cores [5, 6, 10]. As mentioned above, TRUS examination can often identify multiple hypo-echoic areas within the prostate which, in fact, ultimately turn out to be non-cancerous on needle biopsy [1, 3, 4]. However, if mp-MR identifies a highly suspicious prostate lesion, fusion of that stored MR image with the real-time TRUS image may indicate which specific hypo-echoic TRUS lesion best corresponds with the suspected lesion on mp-MR, and should therefore be targeted for TRUS-biopsy. Novel computer-assisted image processing systems may further enhance biopsy accuracy [11–14].

Yet, it must be remembered that all MR-TRUS image-fusion biopsies are ultimately guided by, and performed under, real-time TRUS imaging. It therefore follows that whether or not the MR-suspected lesion is also visible on TRUS could impact the diagnostic yield of image-fusion-guided biopsies. This issue is of considerable practical importance, but has not been elucidated till date. In this study, we evaluate the clinical utility of MR-TRUS elastic image-fusion guidance for targeted prostate biopsy. We also compare diagnostic yield of image-targeted biopsies for prostate lesions that are suspicious only on mp-MR versus lesions that are suspicious on both mp-MR and TRUS.

Materials and methods

Between January 2010 and September 2013, after obtaining Institutional Review Board approval, we retrospectively identified 167 consecutive patients without prior diagnosis of prostate cancer who underwent pre-biopsy mp-MR for a clinical indication for prostate biopsy. Of these, we excluded 40 patients in whom the MR did not identify any suspicious lesions for clinically significant cancer. This resulted in a cohort of 127 patients in whom the MR identified a lesion that was highly (n = 50) or likely (n = 77) suspicious for clinically significant cancer. A flow chart of the number of men who were suitable for study inclusion is presented in Fig. 1. Clinically significant prostate cancer (sPCa) was defined as biopsy Gleason score≥ 7 and/or maximum cancer core length ≥5 mm [14, 15].

Fig. 1.

Schematic tree of study cohort: a flow chart of the number of men who were suitable for study inclusion is presented

MR imaging and MR/TRUS fusion

Mp-MR was performed using a 3 Tesla scanner with pelvic-phased array coils at least 1 day before prostate biopsy. Apparent diffusion co-efficiency (ADC)-map in diffusion weighted images had the same orientation as transverse T2 weighted (T2-w) images. Dynamic contrast-enhanced (DCE) image data were post-processed with pharmacokinetic analysis software (iCAD, Nashua, New Hampshire) to estimate tissue physiological parameters, including transfer constant (Ktrans) and efflux rate constant (kep). Lesions suspicious for clinically significant cancer in T2w, ADC-map, iCAD-DCE were scaled on a 3-point score by an experienced radiologist blinded to the TRUS findings (SP): highly suspicious or likely suspicious for clinically significant cancer or unlikely for clinically significant cancer. For patients with multiple MR lesions, the index MR lesion was assigned by the higher suspicious scale per patient and analyzed in the study. Real-time TRUS images were obtained using an end-firing probe (3D5–9EK or 3D4–9ES 3D-volume-probe, Accuvix-V10 US-machine, Samsung Medison America, Cypress, CA). On the day of prostate biopsy, an experienced urologist (OU), blinded to the MR findings, first assessed the prostate with multi-parametric TRUS (including gray scale and Doppler) to score the TRUS suspicious lesions on a 3-point score similar to the MR scoring system: highly suspicious (suspicious hypo-echoic lesion with focal increase in Doppler signal), likely suspicious (suspicious hypo-echoic lesion without focal increase in Doppler signal), or unlikely (equivocal hypo-echoic) for clinically significant cancer.

At biopsy, 3D volume data from mp-MR and real-time TRUS images were visualized on the screen of a computer workstation (Urostation®, Koelis, La Tranche, France). MR images of the entire prostate gland and suspicious lesions were semi-automatically contoured onto the 3D TRUS image to achieve elastic (non-rigid) image-fusion. Since real-time 3D TRUS imaging can track prostate movement or deformation, any prostate movement at the time of biopsy was imaged on TRUS and fused with the 3D MR volume data.

Prostate biopsy

All biopsies were performed under local anesthesia in the outpatient clinic by an experienced urologist (OU). All patients first underwent 10–12-core systematic biopsies (SB) followed immediately by image-targeted biopsies (TB) using MR/TRUS fusion image and/or TRUS guidance. All MR or TRUS suspicious lesions underwent at least 1 TB per lesion. When more than 1 core of TB was planned, number of SB cores was limited to 10. Before firing the actual TRUS-guided biopsy needle, a virtual 3D-simulated biopsy trajectory was created and superimposed onto the fusion image in order to confirm trajectory accuracy of the proposed biopsy. This virtual targeting simulation was repeated until the ideal biopsy trajectory, penetrating through the center of the suspicious lesion, was achieved. Only then was the actual biopsy needle fired into the prostate. Immediately after firing the biopsy needle, the needle was maintained in situ within the prostate for the next 3 s, during which time repeat 3D TRUS images were acquired to record the actual needle trajectory. Within seconds, a digitalized version of the trajectory was automatically displayed and overlaid onto the MR/TRUS fused image in the workstation, thus confirming precision of the needle biopsy through the MR-visible lesion.

Histological data of any cancer as well as clinically significant cancer in systematic and targeted biopsies were analyzed. Statistical analyses were performed using SAS-Version 9.2 (SAS Institute, NC) software. Correlations between continuous variables were made using nonparametric Spearman’s rank correlation. All tests were two-tailed, and considered significant if p < 0.05.

Results

Median patient age was 66 years, median PSA was 5.8 ng/ ml and clinical stage was T1c in 108 patients (85 %) and T2a in 19 (15 %) (Table 1). Mean number of MR-suspicious lesions per patient was 1.4 (range 1–3). Mean number of cores per patient was 11 for systematic biopsies and 2.8 for MR-TRUS fusion-targeted biopsies. Spatial location of each biopsy trajectory was documented with TRUS-tracking software, thus creating a digitalized 3D map of each biopsy-proven cancer lesion (Figs. 2, 3).

Table 1.

Patient characteristics

Biopsy history	Number (%)
No prior prostate biopsy	57 (45 %)
Prior biopsy negative for cancer	70 (55 %)
PSA median (range)	5.8 ng/ml (1.4–28.8)
Age	66 years (39–81)
DRE, clinical stage, T1c/T2a	108 (85 %)/19 (15 %)
MRI suspicious level Highly suspicious/Likely	50 (39 %)/77 (61 %)
TRUS suspicious level Highly suspicious/Likely/Unlikely	52 (41 %)/67 (53 %)/8 (6 %)

DRE digital rectal examination, TRUS trans-rectal ultrasound

Fig. 2.

Case 1: a 61-year-old man with PSA 11.1 ng/ml had undergone two prior negative extended biopsies over the preceding 2 years at an outside facility. mpMR (upper figures) suggested a highly suspicious lesion in the left-anterior mid prostate, which corresponded with a hypo-echoic lesion on gray-scale TRUS (left bottom figure, with its diagrammatic representation in the middle bottom panel). We performed 3D MR/TRUS fusion targeted biopsy (right bottom figure) using a specific function of the Urostation® (Koelis) which displays an an over-laid ‘virtual biopsy trajectory’ onto the 3D MR/US fusion image, thus allowing confirmation of the correct biopsy trajectory. Actual targeted biopsy was performed immediately thereafter, without any change in the direction of TRUS needle, to precisely sample the lesion. Biopsy revealed Gleason 6 (3 + 3) cancer, 12 mm cancer core length (75 % of core). Mp-MR multi-parametric magnetic resonance, TRUS trans-rectal ultrasound

Fig. 3.

Case 2: a 74-year-old man with PSA 8.6 ng/ml and stage T1a Gleason 5(3 + 2) cancer (3–4 % cancer involved in the chips of a TUR-P in 2007) had already undergone three sets of negative extended-pattern biopsy at an outside facility. mpMR (upper figures) suggested a highly suspicious lesion in the left-anterior lateral transition zone and/or anterior horn peripheral zone, which corresponded with a hypo-echoic lesion on gray-scale TRUS (left bottom figure, with its diagrammatic representation in the middle bottom panel). MR/TRUS fusion targeted biopsy upgraded the lesion to Gleason 7 (3 + 4) cancer with 14 mm cancer core length (90 % core). Based on these findings, the patient terminated active surveillance and crossed-over to radical prostatectomy. Mp-MR multi-parametric magnetic resonance, TUR-P transurethral resection of prostate, TRUS trans-rectal ultrasound

In the overall cohort (n = 127), biopsies were positive for any cancer and clinically significant cancer in 78 (61 %) and 56 (44 %) patients, respectively. Targeted biopsies outperformed systematic biopsies as regards detection of any cancer (61 vs. 41 %, p = 0.007), detection of clinically significant cancer (43 vs. 23 %; p = 0.0013), cancer core length (7.5 mm vs. 3.9 mm; p = 0.0002), and positive rate per core (56 vs. 12 %; p < 0.0001), respectively (Table 2).

Table 2.

Comparison between systematic and targeted biopsy

	Systematic random biopsy	Targeted biopsy	p value
Number of core (per patient)	11.0 (10–12)	2.78 (1–6)	<0.0001a
Positive for cancer per core	165/1,398 (11.8 %)	198/354 (55.9 %)	<0.0001c
Cancer core length (mm), median (range)	3.9 mm (0.2–8.5)	7.5 mm (0.5–18)	0.0002a
Positive for any cancer	52/127 (40.9 %)	78/127 (61.4 %)	0.007c
Gleason score of 6/7/8/9/10 (highest per patient)	30/15/3/3/1	33/35/4/4/2	0.7b
Positive for clinically significant cancer	29/127 (23 %)	54/127 (43 %)	0.0013c

^*Criteria of biopsy-based clinically significant cancer lesion: Gleason score ≥ 7 and/or maximum cancer core length ≥ 5 mms

^aMann–Whitney’s U test

^bSpearman’s correlation coefficient by rank test

^cChi-square test

‘Highly suspicious’ lesions on mp-MR better predicted greater cancer burden compared to ‘likely suspicious’ lesions on mp-MR for any cancer (90 vs. 34 %; p < 0.0001), clinically significant cancer (76 vs. 21 %; p < 0.0001) and cancer core length (8.9 vs. 5.3 mm; p < 0.0001) (Table 3).

Table 3.

Pre-biopsy MRI suspicion level: correlation with biopsy outcomes

	MRI highly suspicious (HS) for sPCa, n = 50	MRI likely (L) sPCa, n = 77	p value
Positive rate per core	148/180 (82 %)	50/174 (29 %)	<0.0001a
Cancer core length, medians (range)	8.9 mm (0–18)c	5.3 mm (0–14)	<0.0001b
Positive for any cancer	45/50 (90 %)	26/77 (34 %)	<0.0001a
Gleason score of 6/7/8/9/10	22/24/3/2/1	11/11/1/2/1	0.018c
Positive for significant prostate cancer	38/50 (76 %)	16/77 (21 %)	<0.0001a

MRI magnetic resonance imaging sPCa, clinically significant cancer

^aChi-square test

^bMann–Whitney’s U test

^cSpearman’s correlation coefficient by rank test

MR suspicion level significantly correlated with TRUS suspicion level (p < 0.0001): 78 % of ‘highly suspicious’ MR lesions were also deemed ‘highly suspicious’ on TRUS, similarly 73 % of ‘likely suspicious’ MR lesions were deemed so on TRUS (Table 4a). Targeted biopsy of any mpMR-visible lesion that was also ‘highly suspicious’ on TRUS had a significantly higher detection rate for any cancer (92 vs. 45 %; p < 0.0001), for clinically significant cancer (75 vs. 22 %, p < 0.0001), and ‘per core’ positive rate (82 vs. 30 %, p < 0.0001) compared to cases where the lesion was only deemed ‘likely suspicious’ on TRUS (Table 4b).

Table 4.

Impact of corresponding findings of TRUS in MR fusion targeted biopsy

	TRUS highly suspicious (n = 52)	TRUS likely (n = 67)	TRUS unlikely (n = 8)
(a)
MR highly suspicious (n = 50)	39 (78 %)	11 (22 %)	0 (0 %)
MR likely (n = 77)	13 (17 %)	56 (73 %)	8 (10 %)

	MR/TRUS fusion Targeted Biopsy
	TRUS highly suspicious (n = 52)	TRUS likely (n = 67)	TRUS unlikely (n = 8)	p value
(b)
Positive rate per core	150/182 (82 %)	48/158 (30 %)	0/14 (0 %)	<0.0001*
Cancer core length, median (range)	8.6 mm (0–18)	5.3 mm (0–14)	NA	0.01§
Positive for any cancer	48/52 (92 %)	30/67 (45 %)	0/8 (0 %)	<0.0001*
Gleason score (6/7/8/9/10)	17/22/3/4/2	16/13/1/0/0	NA	0.3†
Positive for significant prostate cancer	39/52 (75 %)	15/67 (22 %)	NA	<0.0001‡

p < 0.0001 (Spearman’s correlation coefficient by rank test)

TRUS trans-rectal ultrasound, MRI magnetic resonance imaging

^*Kruskal–Wallis Test

^§Mann–Whitney’s U test

^†Spearman’s correlation coefficient by rank test

^‡Chi-square test

Discussion

Practically speaking, MR-TRUS image-fusion biopsies are performed under real-time TRUS imaging. MR helps select which among the multiple TRUS-visible hypo-echoic lesions is most suspicious, and therefore should be targeted for biopsy. Stated another way, TRUS-visibility of the MR-suspicious lesion allows the biopsy to be guided with greater accuracy, leading to superior detection of clinically significant cancer. Thus, expertise in performing TRUS, as well as in interpreting real-time TRUS and mp-MR, is essential when using MR-TRUS fusion technology.

We found that MR-TRUS fusion-targeted biopsies better identified clinically significant cancer compared to systematic biopsies. The higher the MRI suspicion score, the greater was the likelihood for detecting clinically significant cancer. Furthermore, when a MR-suspicious lesion is also visible on real-time TRUS, biopsy targeting becomes more accurate, increasing the detection rate of clinically significant cancer. Conversely, MR-TRUS fusion-targeted biopsy is less accurate when the MR-visible lesion is not visualized on TRUS. This is a key finding, which implies that real-time TRUS plays an essential role during MR-TRUS fusion biopsy. The underlying concept is that the fused MR image is inherently a ‘virtual’ image, while the real-time TRUS image is the ‘real’ image. The prostate is a mobile and deformable organ, whose shape may change and/or shift in the time period between obtaining the pre-biopsy MR and performance of the TRUS-guided needle biopsy. It is important to recognize that the pre-biopsy MR presents a ‘virtual’ target that neither shifts nor deforms during actual needle insertion; conversely, TRUS presents the ‘real’ target which does. Since real-time TRUS visualizes both the biopsy needle and the intra-prostatic target in real-time, it is vital for precise MR-TRUS image-fusion. Thus, when performing image-guided prostate biopsy, it is important to evaluate not only the ‘virtual’ fused MR image, but also the real-time TRUS image.

Image-fusion between MR and TRUS involves multiple steps, including (a) image acquisition of the 3D prostate volume data on both MRI and TRUS, (b) segmentation of the prostate and target contours, (c) image fusion, (d) real-time TRUS guidance of the biopsy and (e) recording of each biopsy trajectory. An error in any one of these steps can lead to sampling error in the MR/TRUS fusion biopsy. When the MR lesion is also visible on real-time TRUS, real-time TRUS can precisely guide the needle to the real lesion, even if the image-fusion process has some errors. However, when the MR lesion is invisible on real-time TRUS, we must rely solely on the MR-derived virtual image of the target; as such, the increased potential for sampling error.

Various benign intra-prostatic lesions can appear as an abnormality on TRUS [3, 4]. As such, TRUS examination may reveal multiple suspicious lesions in each prostate gland, which are inherently non-specific for cancer. The MR-US fusion technique helps identify which specific TRUS-visualized lesion is also suspicious on MR and should therefore be targeted for biopsy from among the multiple TRUS-only suspicious lesions. Interpretation of imaging is in fact operator dependent; however, the combination of two imaging modality with TRUS and MRI for visualizing prostate cancer would enhance the accuracy, compensating the expertise or operator dependency. This would consequently decrease the number of unnecessary biopsies which would have been performed had only TRUS guidance been used. As such, MR-US fusion technology can enhance biopsy targeting accuracy for detecting clinically significant cancer.

We employed certain safeguards to minimize potential errors in image-fusion. First, we used the non-rigid (elastic) fusion technique, which increases targeting accuracy in a deformable soft organ, such as the prostate [11–13]. A recent report indicated that elastic image-fusion outperformed rigid-image fusion as well as the so-called cognitive biopsy [12]. Cognitive biopsy relies on the surgeon’s visual estimation of MR images, without using image-fusion. Since the axial prostate images on MR and the oblique angle of the end-firing TRUS are not in the same axis, it is challenging for the operator to accurately perform cognitive biopsy that requires 3D understanding of the lesion, adjusting for differences in the angle between MR and TRUS images. We believe that 3D MR-TRUS elastic image fusion minimizes such sampling error.

Second, instead of the commonly used 2D TRUS, we used the 3D TRUS probe to record prostate shape and location in real-time, thus enhancing the accuracy of mpMR-TRUS image-fusion. We previously reported the accuracy of 3D TRUS in obtaining real-time 3D prostate volume data [11]. From a mathematical perspective, determining the spatial location of a given geographical point requires availability of three co-ordinates of (x1, y1, z1), which are provided by 3D TRUS. Therefore, 3D TRUS-based image-fusion is more reliable than conventional 2D TRUS. Importantly, our recent study suggested that the real-time 3D TRUS documented biopsy-mapping technology allows a precise re-visiting targeted biopsy of previously documented, low-volume cancer foci with overall accuracy of re-sampling at the site of a known cancer was 86 % per lesion [14].

Our data support prior reports that the higher the MR suspicion score, the greater the likelihood of detecting any cancer and clinically significant cancer [5, 15]. The pre-biopsy mpMR scoring system (Prostate Imaging Reporting and Data System, PI-RADS) developed by the European Society of Uroradiology for prostate MR is in the process of being validated [16, 17]. Although we employed a 3-score system, which we used prior to the more recent 2012 development of PI-RADS, our 3-score system is similar to PI-RADS, wherein highly suspicious corresponds with score 5, likely suspicious with score 4–3 and unlikely with score 2–1 of PI-RADS.

Our study has limitations. First, there is no consensus on biopsy-based definition for clinically significant cancer yet, although we defined as biopsy Gleason score ≥ 7 and/ or maximum cancer core length ≥ 5 mm. We lack correlation of biopsy pathology data regarding lesion size and location, Gleason score, and clinically significant cancer vis-a-vis definitive pathology of the whole-mount prostate gland; these data are a focus of our recent publication [13], in which we reported that MR-TRUS image-fusion guided prostate biopsies can reliably identify location, and primary Gleason pattern of the index tumor lesion in >90 % of patients but have limited predictability of cancer volume, as confirmed by step-sectioned prostatectomy-specimens. Second, when the operator was performing systematic random biopsies, there may have been bias in the random sampling, because the operator may already have prior knowledge of the suspicious lesion based on either TRUS or fused mp-MR images. However, as long as TRUS is used for real-time guidance of prostate biopsy, the operator cannot be completely blind to the TRUS suspicious lesions during systematic sampling. Third, since standardized reporting systems for MR-targeted biopsy studies are only recent developments (START in 2013 [17] and PI-RADS in 2012 [16, 18]), our study lacks reporting of MRI suspicion levels based on these two schema. However, our study suggested that the higher the MRI suspicion score, the more visible on TRUS; and biopsy becomes more accurate, increasing the detection rate of clinically significant cancer. Fourth, all biopsies were performed by an urologist experienced in TRUS, as such these data may not be easily duplicated in general practice. However, when a MR-suspicious lesion for clinically significant cancer is also visible on real-time TRUS, the MR/US fusion-guided biopsy by general urologists without expertise in TRUS could become more accurate, decreasing the sampling error.

Outcomes of MR-TRUS fusion biopsy are dependent on expertise in MR interpretation as well as needle biopsy targeting using the image-fusion system. In order to gain facility with precise MR/US image-fusion as well as precise needle, targeting of MR-suspicious lesions likely has its own inherent learning curve. Every effort to enhance precision in mp-MR interpretation, TRUS interpretation, image-fusion and targeted biopsy techniques (including mutual feedback on imaging and biopsy results between the radiologist and the urologist) is critical for developing a clinically meaningful MR-TRUS fusion biopsy program.

Conclusions

During MR-TRUS image-fusion targeted biopsy, if a MR-suspicious lesion is also visible on TRUS, it increases biopsy accuracy leading to superior detection of clinically significant cancer. Targeted biopsies outperformed systematic biopsies for detection of any cancer and clinically significant cancer. MRI helps select which among the multiple TRUS-visible hypo-echoic lesions is most suspicious for harboring cancer, and therefore a target for biopsy. Expertise in interpretation of both real-time TRUS and mp-MR is essential when using MR-TRUS fusion technology.