[attachment=8806]
A Fast & Automatic Method for Rigid Registration of MR Images of the Brain
ABSTRACT
In this paper a fast and automatic method for 3D rigid registration of magnetic resonance images of the human brain is presented. It is a new advanced technique for the registration of images of the human brain. Image registration is an important problem with several applications in Medical Imaging. It is the process that aligns two or more images in a common reference system of spacial coordinates. It enables comparative analysis of images acquired at different occasions. Intra-subject rigid registration requires a minimal set of parameters to be computed, and is sufficient for organs with no significant movement or deformation, such as the human brain. Rigid registration has also been used as the first step before inter-subject deformable registration. This new technique is based on automatic brain segmentation, automatic mid-sagittal plane location, image alignment, and a greedy transformation.The methodcombines previous approaches for mid-sagittal plane locationand brain segmentation with a greedy-search algorithmto find the best match between source and target images.We evaluated the method on 200 image pairs: 100 withoutstructural abnormalities and 100 with artificially createdlesions, such that it was possible to quantify the registrationerrors. The method achieved very accurate registrationwithin a few seconds.
INTRODUCTION
The development of algorithms for the spatial transformation and registration of tomographic brain images is a key issue in several clinical and basic science medical applications. Several techniques have been introduced for the registration of the magnetic resonance images of the human brain.
Here, Fernanda O. Favretto[1] has proposed a fast and automatic method for 3D rigid registration where the point subsets are obtained from the surface of the brain. Various aspects of this method are discussed in this report. It is achieved through segmentation [2] and their correspondence is found in two steps. The first step aligns the images by the mid-sagittal plane (MSP) [3] and the second step completes registration by using a greedy-search algorithm to find the mapping function between the source (reference) and target point subsets. Indeed, the MSP location already uses the brain segmentation so we are also taking advantage of a sub product of the MSP location approach for registration.
The literature of medical image registration is vast, but this method presents several desirable characteristics simultaneously, which make of it an important contribution: It is fully 3D, simple, fast, and automatic. It has been extensively evaluated on 3D MR-T1 images of the human brain. It is based on automatic brain segmentation, automatic mid-sagittal plane location, image alignment, and a greedy transformation search. The experiments involved 200 target images, 100 without structural abnormalities and 100 with artificially created lesions, such that it was possible to quantify the registration errors.
IMAGE REGISTRATION
Image registration is the process that aligns two or more images in a common reference system of spacial coordinates. This alignment is often done by taking one image domain as reference, transforming corresponding points from each of the other image domains into the reference system, and then extending this transformation to the remaining pixels. The main problems are the choice of suitable point subsets in each image domain and the determination of their mapping functions onto the reference system. We address both problems in the context of rigid registration between 3D magnetic resonance (MR) images of the human brain, obtained from a same individual in different time instants. The goal is to register pre- and post-surgical images from epilepsy patients who had lesion brain tissues removed to eliminate the foci of the seizures. An additional challenge is that, due to tissue removal, some points do not have correspondents in the reference subset. Some patients did not cease the seizures after surgery. Localization and quantification of the variations in the brain tissues for such patients can help neurologists to understand the phenomenon and develop new treatments.
Image registration is acrucial step in all image analysis tasks in whichthe final information is gained from thecombination of various data sources like inimage fusion, change detection, andmultichannel image restoration. Typically,registration is required in remote sensing(multispectral classification, environmentalmonitoring, change detection, weatherforecasting, creating super-resolution images,integrating information into geographicinformation systems), in medicine(combiningcomputer tomography (CT) and NMR data toobtain more complete information about thepatient, monitoring tumour growth, treatmentverification, comparison of the patient’s datawith anatomical atlases), in cartography (mapupdating), and in computer vision (targetlocalization, automatic quality control), to namea few. During the last decades, image acquisitiondevices have undergone rapid development andgrowing amount and diversity of obtainedimages invoked the research on automatic imageregistration.