Personalized modeling of prostate deformation based on elastography for MRI-TRUS registration

Abstract

Fusion of pre- and intra-operative imaging modalities allows the accurate targeting of prostate interventions. We propose a personalized statistical motion model (PSMM) for the deformable registration of prostate magnetic resonance (MR) and 3D transrectal ultrasound (TRUS) images. Comparing to traditional statistical motion model (SMM) methods, for the first time, the biomechanical information obtained from elastography is incorporated into a finite element (FE) model, which is created from preoperative MR images. PSMM is then constructed from training data generated by simulating the prostate deformation from the personalized FE model with realistic boundary conditions. Finally, prostate MR and TRUS images are registered by integrating the PSMM into a robust point matching (RPM) framework. Using data acquired from phantoms, the efficiency of the PSMM and the accuracy of MRI-TRUS registration were evaluated. The mean target registration error was less than 1.7mm, which demonstrates our PSMM can achieve more physically plausible deformation estimation and accurate registration.