A Novel and Fully Automated Registration Method for Prostate Cancer Detection Using Multiparametric Magnetic Resonance Imaging

Background and Objective: Multiparametric (mp)-Magnetic Resonance Imaging (MRI) is emerging as a powerful test to diagnose and stage prostate cancer (PCa). However, its interpretation is a time consuming and complex feat, requiring dedicated radiologists. Computer-aided diagnosis (CAD) tools could allow better integration of data deriving from the different MRI sequences in order to obtain accurate, reproducible, non-operator dependent information useful to identify and stage PCa. Unfortunately, due to the differences between MRI scanning protocols and some degree of patient movement and image deformation, CAD output may be inaccurate. This study tests the improvements in terms of PCa detection of a CAD system, derived from the application of automatic algorithms to register dynamic contrast enhanced (DCE)-MRI volumes and diffusion weighted (DW) images to T2-weighted (T2-w) images. Methods: A fully automatic 3D algorithm to register DCE-MRI volumes and DW images to T2-w images is applied. First, a 3D rigid registration method between DCE and T2-w images was developed to correct for patients movements. Then, a non-rigid registration method was created to correct misalignment between T2-w and DW images, mainly due to image distortion and patients movements. To test for improvement, several measurements were implemented on 20 patients, based on both the distances between anatomical landmarks and the effect on a previously presented CAD system. Results: Results showed a mean distance of about 1 mm between landmarks after the registration for both DW/T2-w and DCE/T2-w algorithms, thus correcting respectively 74% and 43% of the initial displacement. Besides, the advantages of bringing the method into clinical application have been supported by the 19% increase of the performances of CAD system. Conclusions: The application of a fully automatic registration framework allows high quality registration of different MR sequences and improves pixel-by-pixel detection of tumoural tissue within the prostate gland. Initial results on the implementation of the framework in the CAD pipeline are promising.