Elastic image registration of MR/TRUS for targeted prostate biopsy

摘要

In order to provide prostate biopsy operators with information from pre-operative MR images, MR images must be transformed accurately to the Transrectal Ultrasound (TRUS) real-time images. In our work, we used the well-known Robust Points Matching (RPM) algorithm to find the corresponding point pairs from the prostate contours of MR and TRUS and the thin plate splines (TPS) to transform MRI to TRUS images elastically. Compared with choosing landmarks based on key structures adopted by Nasr Makni[11], we find generating point sets by down-sampling the manual delineated contours in two modality images is much more robust and easier, as it depends little on the experience of operators as well as the quality of clinical ultrasound imaging. Using data acquired from 5 patients, the mean DSC (Disc Similarity Coefficient) we get after performing RPM-TPS registration was 79.43%, compared to the 60.20% got by the manual rigid registration.