Biomechanically Constrained Surface Registration: Application to MR-TRUS Fusion for Prostate Interventions

Abstract

In surface-based registration for image-guided interventions, the presence of missing data can be a significant issue. This often arises with real-time imaging modalities such as ultrasound, where poor contrast can make tissue boundaries difficult to distinguish from surrounding tissue. Missing data poses two challenges: ambiguity in establishing correspondences; and extrapolation of the deformation field to those missing regions. To address these, we present a novel non-rigid registration method. For establishing correspondences, we use a probabilistic framework based on a Gaussian mixture model (GMM) that treats one surface as a potentially partial observation. To extrapolate and constrain the deformation field, we incorporate biomechanical prior knowledge in the form of a finite element model (FEM). We validate the algorithm, referred to as GMM-FEM, in the context of prostate interventions. Our method leads to a significant reduction in target registration error (TRE) compared to similar state-of-the-art registration algorithms in the case of missing data up to 30%, with a mean TRE of 2.6 mm. The method also performs well when full segmentations are available, leading to TREs that are comparable to or better than other surface-based techniques. We also analyze robustness of our approach, showing that GMM-FEM is a practical and reliable solution for surface-based registration.

Learning Non-rigid Deformations for Robust, Constrained Point-based Registration in Image-Guided MR-TRUS Prostate Intervention

“…The non-rigid FFD transformation with high degrees of freedom (DoFs) offered too much flexibility and the algorithm incorrectly registered the prostate surface to segmentation noise. Similarly, other studies have reported increased TRE for unconstrained non-rigid registration methods over rigid registration (Cool et al, 2011; Khallaghi et al, 2015b) and that unconstrained, deformable registration did not show significant improvements over affine registration (Delongchamps et al, 2013; Fedorov et al, 2015). We have seen analogous results when performing intensity-based multi-modal brain registration with an SDM (Onofrey et al, 2016), where rigid registration can outperform unconstrained non-rigid registration due to noisy data.…”

“…Localizing these landmarks was challenging due to the ambiguous delineation of the prostate gland surface in parts of the TRUS imaging with little tissue contrast. We found that patient-specific cysts and other “cyst-like” anatomical structures appearing in both the MR and TRUS images were the most reliable to localize, and these types of landmarks have also been used to calculate TRE in other studies (Hu et al, 2012, 2015; Fedorov et al, 2015; Khallaghi et al, 2015b,a). The 29 identified landmarks used in Sec.…”

“…Moradi et al (Moradi et al, 2012) non-rigidly register manually segmented prostate binary image masks, while Fedorov et al (Fedorov et al, 2015) non-rigidly register signed distance maps of manually segmented prostates. Surface-based non-rigid registration methods treat the segmentations as points in 2D slice contours (Reynier et al, 2004; Cool et al, 2011; Mitra et al, 2012a; Zettinig et al, 2015), as points in 3D surfaces (Narayanan et al, 2009; Karnik et al, 2010; Natarajan et al, 2011; Fedorov et al, 2015; Khallaghi et al, 2015b,a; van de Ven et al, 2015; Onofrey et al, 2015b), or as a set of basis functions, for example as spherical harmonics (Moradi et al, 2012). The state-of-the-art, clinical Artemis prostate biopsy system and its ProFuse Bx multimodal image fusion software (Eigen, Grass Valley, CA), relies on intra-procedure, semi-automated TRUS segmentation (Ladak et al, 2000) and a deformable surface registration algorithm to align the prostate surfaces segmented in both the MRI and TRUS images (Narayanan et al, 2009).…”

“…Surface-based registration algorithms such as robust point matching (RPM) (Rangarajan et al, 1997; Chui and Rangarajan, 2003) and coherent point drift (CPD) (Myronenko and Song, 2010) attempt to handle noisy surface data by allowing for soft point correspondences and outliers in the data. MR-TRUS fusion methods have made use of RPM (Khallaghi et al, 2015b; Onofrey et al, 2015b) and CPD (Fedorov et al, 2015; Khallaghi et al, 2015b,a; Zettinig et al, 2015). These methods have been shown to be robust to noisy segmentations with synthetically added perturbations (Onofrey et al, 2015b; Khallaghi et al, 2015a) and to partial segmentations missing large sections of the prostate (Fedorov et al, 2015; Khallaghi et al, 2015b,a).…”

“…A general criticism of surface-based registration methods is that they can only reliably register data that is close the surface itself since there is no information provided to the registration algorithm inside the shape boundary far from the surface. Biomechanical models address this limitation by acting as a regularization constraint on the non-rigid transformation (Moradi et al, 2012; Hu et al, 2012, 2015; Fedorov et al, 2015; Khallaghi et al, 2015b,a; van de Ven et al, 2015). These models treat the prostate as a linear elastic material, and require tuned parameters to best approximate the elastic tissue properties of the gland.…”