Abstract

Medical image registration is a critical image processing task in many applications such as image-guided surgery (IGS) and image-guided radiotherapy. Herein, a novel automatic inter-modal affine registration technique is proposed based on the correlation ratio (CR) similarity metric firstly. The technique is demonstrated through registering intra-operative ultrasound (US) scans with magnetic resonance (MR) images of 22 patients from a publicly available database. By using landmark-based mean target registration errors (mTRE) for evaluation, the technique has achieved a result of 2.79$\pm$1.13 mm from an initial value of 5.40$\pm$4.31 mm. A nonparametric statistical analysis performed using the Wilcoxon rank sum test shows that there is a significant difference between pre- and post-registration mTREs with a $p$-value of $0.0058$. To achieve this result, the MRI was deemed as the fix image ($I_f$) and the US as the moving image ($I_m$) and then $I_m$ was transformed to align with $I_f$. Covariance matrix adaptation evolutionary strategy (CMA-ES) was utilized to find the optimal affine transformation in registration of $I_m$ to $I_f$. In addition to quantitative validation using mTRE, the results were validated qualitatively by overlaying pre- and post-registration US and MRI to allow visual assessment of the alignment. The proposed fully automatic registration method significantly improved the alignment of MRI and US images and can therefore be used to aid neurosurgeons in resection of brain tumors. In addition to proposing new methods for registration of US and MRI, three different datasets of corresponding CT and US images of vertebrae were collected and presented. In the first dataset, two human patients’ lumbar vertebrae are presented and the US images are simulated from the CT images. The second dataset includes corresponding CT and US images of a phantom, made of post-mortem canine cervical and thoracic vertebrae. The third dataset includes the CT and US images of a lamb’s lumbar vertebrae. For the two latter datasets, 15 corresponding landmarks were provided and fiducial registration of the corresponding images was performed to acquire a silver standard ground truth of the registration. This dataset will be released online to allow validation of US-CT registration techniques.