PHSOR12 Presentation Time: 9:55 AM

Abstract

Purpose

Utilizing a female pelvic phantom and following TG-132 methodologies, we developed a novel digital phantom to validate the MR/TRUS contour-based elastic registration algorithm embedded in the commercial needle-tracking software developed by Eigen Health. The reliability of this registration algorithm is crucial for commissioning a pre-existing prostate needle-tracking biopsy system as a gynecological HDR brachytherapy tool.

Methods

A previously optimized female pelvic phantom was embedded with six vitamin D fiducial markers. Four markers were placed on the exterior uterine surface (2 lateral, 1 anterior, 1 superior), and two were placed on the uterine interior (1 central, 1 anterior surface). A T2w MR image set of the phantom was acquired and segmented in 3DSlicer, an open-source image analysis software, to define the uterus and fiducials, which were then exported as STL files. Manual deformations were applied in Rhinoceros, a 3D CAD software, and proprietary scripts converted the deformed contours into raw binary images. A custom MATLAB code converted these images into DICOM format for import into Eigen's ProFuseCAD software which was used to segment the deformed uterus and define the fiducials as regions of interest (ROIs). The uterus contour and fiducial ROIs were then imported into Eigen's needle-tracking system. Using a BK Medical linear side-fire TRUS probe, a 150-degree lateral-to-lateral sweep of the non-deformed phantom was acquired. The uterus was segmented on the resulting 3D TRUS volume. Registration between the MRI and TRUS volumes was performed by first rigidly registering the two volumes in the sagittal and transverse planes based on 2 paired points on the uterus. A surface based elastic registration algorithm was then applied to complete the registration process. Target Registration Error (TRE) was estimated between the automatically marked centroid of each fiducial ROI and the manually marked fiducial centers from visual inspection in the acquired TRUS volume. This process was repeated for six different manual deformations applied to the digital phantom uterus to evaluate the system's registration quality. A control case without deformations was also examined.

Results

All six of the embedded fiducials were successfully visualized on MR and included in the digital phantom. The limited US FOV restricted visibility of the anterior fiducial, making its coordinates unreliable. For the five visible fiducials, the average ΔX, ΔY, ΔZ between the automatically generated point and the manually defined centroid point was 2.29±1.53, 0.94±1.02, 3.65±1.40mm, respectively. The overall average vector displacement of the location was 4.68 ± 1.49 mm.

Conclusion

Our study demonstrates a novel method to validate the deformable registration algorithm used by the Eigen system. Distance results in each plane for all visible fiducials were within the 2-3mm TRE tolerance stated in TG-132. However, the overall displacement vector was commonly greater than the tolerance. There is considerable uncertainty in manually selecting the centroid point of the irregularly shaped fiducials. Further studies are needed to refine the validation process and investigate the sources of uncertainty in the registration results.