Comparison of Elastic and Rigid Registration during Magnetic Resonance Imaging/Ultrasound Fusion-Guided Prostate Biopsy: A Multi-Operator Phantom Study

Abstract

Purpose:

The relative value of rigid or elastic registration during magnetic resonance imaging/ultrasound fusion guided prostate biopsy has been poorly studied. We compared registration errors (the distance between a region of interest and fiducial markers) between rigid and elastic registration during fusion guided prostate biopsy using a prostate phantom model.

Materials and Methods:

Four gold fiducial markers visible on magnetic resonance imaging and ultrasound were placed throughout 1 phantom prostate model. The phantom underwent magnetic resonance imaging and the fiducial markers were labeled as regions of interest. An experienced user and a novice user of fusion guided prostate biopsy targeted regions of interest and then the corresponding fiducial markers on ultrasound after rigid and then elastic registration. Registration errors were compared.

Results:

A total of 224 registration error measurements were recorded. Overall elastic registration did not provide significantly improved registration error over rigid registration (mean ± SD 4.87 ± 3.50 vs 4.11 ± 2.09 mm, p = 0.05). However, lesions near the edge of the phantom showed increased registration errors when using elastic registration (5.70 ± 3.43 vs 3.23 ± 1.68 mm, p = 0.03). Compared to the novice user the experienced user reported decreased registration error with rigid registration (3.25 ± 1.49 vs 4.98 ± 2.10 mm, p <0.01) and elastic registration (3.94 ± 2.61 vs 6.07 ± 4.16 mm, p <0.01).

Conclusions:

We found no difference in registration errors between rigid and elastic registration overall but rigid registration decreased the registration error of targets near the prostate edge. Additionally, operator experience reduced registration errors regardless of the registration method. Therefore, elastic registration algorithms cannot serve as a replacement for attention to detail during the registration process and anatomical landmarks indicating accurate registration when beginning the procedure and before targeting each region of interest.

MAGNETIC resonance imaging and commercially available software assisted MRI/US FBx platforms have increased the detection of csPCa while simultaneously decreasing the diagnosis of indolent disease.^1–3 FBx combines the benefits of MRI (high spatial resolution and pre-identified suspicious lesions) with the versatility of US (high temporal resolution) in an outpatient setting.^4,5 Registration is the often tedious process of bringing separate imaging modalities (MRI and US) into spatial alignment for ease of viewing, either side by side or blended on top of each other using fusion software.⁶ While it is often time-consuming, proper registration and fusion of images are necessary since small targeting errors in the range of 1 to 2 mm may result in missing a visualized csPCa.⁷

Commercial FBx platforms can use different methods of registration (rigid or elastic) to fuse MRI data with US. Rigid registration enables the user to manually rotate and translate the MRI and US images with respect to each other to produce the best global alignment between the images, although the images themselves do not change. This is the simplest method as is does not accommodate deformation of the prostate gland. However, rigid registration maintains the true patient anatomy detected on MRI.⁴

In contrast, elastic registration uses a software algorithm to compensate for changes in the segmented prostate shape, which may occur between the pre-procedural MRI and initial or subsequent intraprocedural imaging during the FBx procedure.^5,8–12 The user may still rotate or translate the rigid images initially to create alignment but after elastic transformation is applied the semiautomated software algorithm is permitted to stretch and deform the segmented margin of the MRI to better fit the segmented real-time US margin or vice depending on the platform. This is done to maximize fusion image alignment and create what seems like a more precise depiction of the lesion superimposed on US but it may in fact warp or distort the true anatomy in the process.⁴

While there are several different elastic registration algorithms,¹³ conceptually one can think of the general process as a set of connected pins (segmented features on MRI) being inserted in a corresponding set of holes (segmented features on US). Under a rigid registration framework these pins would be connected by a stiff and rigid material, forcing them to move in unison. With an elastic registration framework these pins would be connected by an elastic rubber band, allowing them to deviate from the original configuration and deform to best fit the set of holes.

Prior groups have compared elastic and rigid registration in the experimental or clinical settings with conflicting results.^14–19 Experimental and clinical studies have some limitations. Studies performed under experimental conditions are well controlled but lack realism while clinical studies often lack final pathology results for comparison. Additionally, clinical studies introduce additional variables such as tumor heterogeneity, MRI scoring systems, definitions of csPCa, patient inclusion criteria, user experience levels and different FBx systems such as differences in elastic registration algorithms.^14,19

We directly compared rigid and elastic REs in a controlled setting using a phantom model prostate. A standard clinical workflow was mimicked with MRI acquisition, radiologist interpretation, registration on a commercial platform and urologist FBx²⁰ but controlling for confounding clinical factors. Secondary aims included determining whether RE differences exist in the user experience level, the target location or the ultrasound plane between the 2 registration methods.

MATERIALS AND METHODS

This study was performed on a single phantom prostate model and did not involve human subjects. One highly experienced user who had experience with more than 1,000 FBx procedures and 1 novice who had experience with fewer than 5 FBx procedures participated in this study.

Four gold fiducial markers (CIVCO, Coralville, Iowa) visible on MRI and US were placed throughout the Model 070L prostate phantom (CIRS, Norfolk, Virginia) which was then imaged by a 3 Tesla Achieva MRI scanner (Philips, Bothell, Washington) using a BPX-30 endorectal coil (Medrad®). A radiologist segmented the prostate contour on MRI and labeled fiducial markers as ROIs 1 to 4 (fig. 1). The model prostate margins and the ROI target information were superimposed on anatomical T2-weighted MRI scans and transferred to the FBx workstation for the study using a DynaCAD workstation (Invivo, Gainesville, Florida).

Figure 1.

T2-weighted sagittal image of prostate phantom (a). Arrowhead indicates ROI representing lesion location in prostatic apex. T2-weighted axial image of prostate phantom of same lesion shows artifacts from previous biopsy needle tracts (b). Asterisks indicate phantom deformation. Dagger represents bladder. Green outline indicates prostate for purpose of MRI image co-registration with real-time intraprocedural US.

The experienced and the novice FBx users performed experimental biopsies using the UroNav FBx platform (Invivo) (fig. 2). Each trial first consisted of a US 3-dimensional volume reconstruction of the model prostate by sweeping with a C9–5 transducer (Philips) from base to apex. The FBx users next segmented the phantom edge on US. The platform then calculated the registration algorithms and fused the MRI and US scans. Finally the users targeted ROIs 1 to 4, which were marked on the FBx, and then targeted the corresponding fiducial on US but only after rigid and elastic registration at separate sessions (figs. 3 to 5). Each ROI and each US only fiducial marker was targeted with 1 axial and 1 sagittal US view for a total of 16 RE distances per trial for comparison. RE was defined as the measured distance between the ROI and the fiducial markers on the FBx platform.

Figure 2.

Experimental biopsy equipment setup, including UroNav FBx platform (A), US transducer (B), electromagnetic tracker (C) and phantom prostate model (D) containing 4 gold fiducial markers visible on MRI and US.

Figure 3.

Intra-experimental screen capture shows rigid registration ROI 1 with axial view targeting. MRI/US fusion image demonstrates good image registration (a). Note fiducial marker shadow visible behind corresponding ROI target (B). Contoured MRI prostate data (C, pink outline) corresponds to contoured prostatic capsule (fig. 1). Visible phantom MRI data shows contoured prostate (C) and ROI 1 (E) with(exact coordinates of targeted locations collected with electromagnetic tracking system (d).

Figure 5.

Intra-experimental screen capture shows axial view of elastic registration of ROI 2 (a). Fusion guided biopsy user targets ROI 2 corresponding fiducial marker (B) by aligning it with needle guide (pink dotted line). Electromagnetic tracking system collects targeted location coordinates (b). RE is calculated by measuring distance between previously targeted ROI coordinates (fig. 4) and those of corresponding fiducial marker.

Seven repeat FBxs were needed to power the study to detect a RE difference of 2.5 mm at 80% power and a 5% significance level. RE distances between the rigid and elastic registrations were compared with the paired t-test. The secondary outcomes were RE measured distances for the experienced and the novice users, ROI locations, and axial vs sagittal US planes, which were also compared by the paired t-test. Rigid and elastic registration variances were compared with the 1-sided F test at a 5% significance level. Statistics were performed with Stata/SE™ 15.

RESULTS

A total of 224 RE measurements were recorded, including 112 each for the experienced and the novice FBx users. Table 1 shows a summary of RE measurement results by user experience level, ROI location and US view orientation. Distances from the fiducial markers to the model edge were measured on MRI in the axial and sagittal views (table 2 and fig. 6). ROI 4 (edge lesion), simulating an edge lesion in the prostate gland, was closest to the model edge while ROIs 1 to 3 (central lesions) were centrally located.

Table 1.

Mean registration error measurements

		Mean ± SD Registration Error (mm)
	No. Pts	Rigid	Elastic
Overall	224	4.11 ± 2.09	4.87 ± 3.50
User:
Experienced	112	3.25 ± 1.49	3.94 ± 2.61
Novice	112	4.98 ± 2.10	6.07 ± 4.16
Lesions:
Edge	28	3.23 ± 1.58	5.70 ± 3.43
Central	84	3.23 ± 1.46	3.36 ± 2.00
Plane:
Axial	112	3.38 ± 1.64	4.32 ± 2.86
Sagittal	112	3.08 ± 1.36	3.57 ± 2.33

Table 2.

Distance from fiducial markers to prostate model edge

		Lesion-Prostate Edge Distance (mm)
Region of Interest	Lesion	Axial	Sagittal	Shortest
1	Central	16.1	15.6	15.6
2	Central	24.8	25.3	24.8
3	Central	17.9	18.5	17.9
4	Edge	16.0	1.7	1.7

Figure 6.

Measuring distances from prostate model edge to fiducial marker labeled ROI 1 in axial (a) and sagittal (b) planes acquired on MRI. Due to distances measured to model edge (green outline) lesion was termed centrally located. This pattern was followed for ROIs 2 and 3. ROI 4 in axial (c) and sagittal (d) views was located closer to model edge in sagittal view and was termed edge lesion.

Overall there was no significant difference between rigid and elastic REs (mean 4.11 ± 2.09 and 4.87 ± 3.50 mm, respectively, p = 0.05). This remained true for the experienced user (mean 3.25 ± 1.49 vs 3.94 ± 2.61 mm, p = 0.09) and the novice user (4.98 ± 2.10 vs 6.07 ± 4.16mm, p = 0.08). The mean RE was significantly lower for the experienced user than for the novice user with rigid registration (3.25 ± 1.49 vs 4.98 ± 2.10 mm) and elastic registration (3.94 ± 2.61 vs 6.07 ± 4.16 mm, each p <0.01). For the experienced user the mean RE of the model edge lesion (ROI 4) was significantly larger when using elastic vs rigid registration (5.70 ± 3.43 vs 3.23 ± 1.68 mm, p = 0.03).

A comparison of elastic and rigid REs in the central lesions did not differ (mean 3.36 ± 2.00 vs 3.23 ± 1.46 mm, p = 0.74). There was no statistical difference in the RE between rigid and elastic registration in the axial plane (mean 3.38 ± 1.64 vs 4.32 ± 2.86 mm, p = 0.14) vs the sagittal plane (3.08 ± 1.36 vs 3.57 ± 2.33 mm, p = 0.35). The rigid registration variance was significantly lower than the elastic registration variance overall (mean 4.37 vs 12.25, F = 0.36 < 0.73), and between the experienced user (2.22 vs 6.81, F = 0.33 < 0.64) and the novice user (4.41 vs 17.30, F = 0.26 < 0.64). Figure 7 summarizes these results.

Figure 7.

Summarized statistical results of rigid (blue bars) vs elastic (orange bars) RE total and experience level (a), and location and US view (b).

DISCUSSION

Elastic registration for FBx was introduced because there was often a visible mismatch between MRI and US images using rigid registration. It was assumed that because registration looked better with elastic registration, it would result in superior results with reduced relative errors. Our data indicate that rigid and elastic registration produced approximately equivalent results, although rigid registration may be slightly superior to elastic registration in terms of reducing variance. Our data also indicate that rigid registration produced a significantly lower RE when targeting ROIs close to the model edge (prostate glandular edge equivalent). In this area the elastic registration algorithm is permitted to warp images to create a more pleasing fusion image but this may occur at the cost of accuracy. Lastly, neither rigid nor elastic registration was superior in the sagittal or the axial plane of view.

Prior groups have compared elastic and rigid registration with conflicting results. Experimental studies have tended to favor elastic registration while clinical studies have shown no significant difference.^14–18 Elastic registration can compensate for changes to the segmented prostate shape, such as those due to the endorectal coil, supine MRI vs decubitus US, force applied by the transducer probe, level, biopsy gun needle entry or operator differences.^5,8–10 It seems intuitive that elastic registration might be superior to rigid registration to reduce the RE as it attempts to account for deformations to the prostate.

Several different methods of elastic and rigid registration have been developed and tested in a highly controlled experimental setting with an elastic RE of 1 to 3 mm as an improvement over rigid RE.^{12,15–18,21,22} An earlier phantom study by Xu et al, which was performed on an experimental platform similar to the commercial platform used in this study, revealed that the overall mean accuracy of this system with the addition of simulated motion was 2.4 ± 1.2 mm.²⁰ All of these results are lower than our RE average. However, our study included a clinical workflow that may have introduced more error.

A single meta-analysis compared rigid and elastic registration in the clinical setting by calculating and comparing the detection OR as a surrogate for the accuracy of registration. No significant difference in the cancer OR was detected for elastic registration vs rigid registration (11 vs 10).¹⁴ In a single study registration methods were indirectly compared by determining the csPCa detection rates of FBx and random biopsy done at the same time in rigid and elastic registration cohorts but again there was no significant difference (p = 0.13).¹⁹ This indicates that circumstances in a completely artificial experimental setting may not reflect what urologists exper during FBx on a commercial platform in the clinic, which is supported in the existing literature.²³ Our data indicate that neither rigid nor elastic registration was superior, lending support to the clinical studies and indicating that user preference may dictate the registration method applied in clinical practice.

Lesion location may have a role in the RE and the registration method. A previous study demonstrated that elastic registration reduced the RE in all prostate zones, especially in the peripheral zone, where the most deformation occurs due to transducer pressure.²⁴ Other studies showed that the RE is affected by lesion location, specifically that it is worse along the probe axis where pressure is applied or in the anterior and base where differences in the MRI and US exploration planes are greatest.^23,25

Our data may add to these findings, indicating that the RE is increased when using elastic registration to target lesions in the periphery of the gland close to the prostate edge. The elastic registration algorithm uses the previously contoured glandular edge on MRI and the intraprocedurally contoured edge on US as the feature to match. This process especially warps the prostate edges to produce a better fit while maintaining capsule smoothness.²¹ The operator tends to want to create a perfectly united fusion image and may not consider the effect that this has on accuracy.

Elastic registration algorithms attempt to match the edges in an equally weighted global fashion. Therefore, if 1 edge is more different than another, this might introduce a larger degree of global correction affecting 1 edge proportionally greater than other edges. This correction may move a target in a way that no longer represents the true location of the lesion. In contrast, rigid registration maintains the integrity of the actual anatomical relationships. Therefore, the operator can adjust a rigid registration by cognitively weighting the edges near a target more iencethan the distal edges far away from the target. Itis possible that rigid registration with a manually weighted edge near an edge target is more accurate than its elastic counterpart. This finding emphasizes that elastic registration is not a replacement for accurately aligning the MRI and US images at the beginning of the procedure and correcting this registration throughout the procedure, particularly just before the next targeted biopsy needle.

Standardization and reproducibility are improved by less variance. Our finding that the elastic registration variance may be higher than the rigid registration variance was found in a prior report.²¹ The semi-autonomous nature of the elastic registration algorithm may unnecessarily deform the MRI scans at the expense of accuracy. Particular care should be taken to minimize gland deformation during the US sweep and when manually matching the edges because discrepancies in MRI or US segmentation may not be corrected by the elastic registration algorithm and in peripheral lesions it may be amplified. Our data show that user experience can reduce the RE of each registration method. This finding helps support the idea that with experience FBx users improve with time.¹⁰

This study has several limitations. FBx users were not blinded to the registration method that they applied during the trials, which could have introduced bias. Additionally, it is difficult to mimic human tissue characteristics on a phantom prostate. Only 1 of many FBx platforms was used and the generalizability of our results may be restricted to this device since elastic registration algorithms differ slightly among platforms. Lastly, when rigid registration and elastic registration are each available intraprocedurally, clinicians can switch between the registration methods depending on which one produces the more accurate match on a per lesion basis. Since we only measured rigid or elastic registration and not each method, our study does not represent this practice trend.

CONCLUSIONS

Elastic registration warps images to create a more pleasing picture to the clinician. However, to our knowledge whether this translates into better targeting during clinical use is unknown. In this phantom model elastic registration did not provide an advantage over rigid registration but increased user experience level reduced the RE of each registration method. Regions of interest closer to the prostate edge may experience significantly increased RE when using elastic instead of rigid registration due to increased distortion of the image closer to the contoured edges of the prostate. As FBx increasingly becomes part of standard urology practice it is important to use best practices when aligning MRI and US images. Although elastic registration produces a fusion image with more overlap, it should not be assumed that elastic registration improves biopsy accuracy.

Figure 4.

Intra-experimental screen capture demonstrates axial view of elastic registration ROI 2. In MRI/US fusion image fusion guided biopsy user targets ROI 2 (B) by aligning ROI 2 with needle guide (pink dotted line) (a). Note ROI 2 corresponding fiducial marker (C) visible in same imaging plane. Electromagnetic tracking system collects coordinates of targeted location ROI 2 (b).

Abbreviations and Acronyms

csPCa

clinically significant prostate cancer

FBx

fusion guided prostate biopsy

MRI

magnetic resonance imaging

RE

registration error

ROI

region of interest

US

ultrasound