Date Published: March 15, 2019
Publisher: Public Library of Science
Author(s): Michael Figl, Rainer Hoffmann, Marcus Kaar, Johann Hummel, Qinghui Zhang.
US image registration is an important task e.g. in Computer Aided Surgery. Due to tissue deformation occurring between pre-operative and interventional images often deformable registration is necessary. We present a registration method focused on surface structures (i.e. saliencies) of soft tissues like organ capsules or vessels. The main concept follows the idea of representative landmarks (so called leading points). These landmarks represent saliencies in each image in a certain region of interest. The determination of deformation was based on a geometric model assuming that saliencies could locally be described by planes. These planes were calculated from the landmarks using two dimensional linear regression. Once corresponding regions in both images were found, a displacement vector field representing the local deformation was computed. Finally, the deformed image was warped to match the pre-operative image. For error calculation we used a phantom representing the urinary bladder and the prostate. The phantom could be deformed to mimic tissue deformation. Error calculation was done using corresponding landmarks in both images. The resulting target registration error of this procedure amounted to 1.63 mm. With respect to patient data a full deformable registration was performed on two 3D-US images of the abdomen. The resulting mean distance error was 2.10 ± 0.66 mm compared to an error of 2.75 ± 0.57 mm from a simple rigid registration. A two-sided paired t-test showed a p-value < 0.001. We conclude that the method improves the results of the rigid registration considerably. Provided an appropriate choice of the filter there are many possible fields of applications.
Ultrasound imaging (US) is real-time, non-invasive and less expensive than many other clinical imaging methods. However, the images are often noisy, contain several typical artefacts and are affected by soft-tissue deformation caused by the application of the US-scan itself. All these factors can contribute to difficulties in the comparison of images of the same object at different times or from different positions i.e. the registration of US images . In order to improve monitoring and diagnostics using standard US technologies there is a certain demand for finding better ways to compare and fuse corresponding images.
The work flow of the complete registration process including the calculation of the TRE is shown in Fig 1. It was divided into four main parts.
Our multi-resolution approach has shown significant improvements of the TRE in both evaluations, the phantom and the patient study. Furthermore, the errors found were in good accordance with  where registrations between liver images were performed. Due to the higher contrast in some parts of the images, we were able to evaluate more target points compared to . In , the accuracy and variability of rigid and non-rigid registrations of transrectal 3D-US images of the prostate were evaluated. Different surface- and intensity-based rigid and nonrigid registration algorithms based on thin-plate splines and B-splines were evaluated. The pre-registration TRE was 7.36 ± 4.17 mm compared to a TRE of 1.96 ± 0.85 mm after non-rigid registration. Nevertheless, no significant difference between rigid and non-rigid registration was found. In contrast, De Silvas et al.  were able to reduce the TRE by 4.75 mm compensating prostate motion induced by the biopsy procedure. Their intensity-based 2D-3D rigid registration algorithm optimized the normalized cross-correlation metric using Powell’s method. Rivas et al.  applied a non-rigid US registration on brain images where a deformation occurred due to a resection of brain tumors. The deformation was modeled with free-form cubic B-splines. Their registration algorithm reduced the mean TRE from an initial value of 3.7 mm to 1.5 mm. A similar approach related to the presented registration method is given by the determination of geometric moment invariants for each point in the image [18, 19] where attribute vectors are determined from local spatial intensity histograms. Their method was applied to magnetic resonance (MR) images and has proven to be robust and reliable. The transfer of this method to US images would be challenging due to the huge amount of speckle artifacts.