Fast Nonsupervised 3D Registration of PET and MR Images of the Brain

Abstract

Summary:We propose a fully nonsupervised methodol ogy dedicated to the fast registration of positron emission tomography (PET) and magnetic resonance images of the brain. First, discrete representations of the surfaces of interest (head or brain surface) are automatically ex tracted from both images. Then, a shape-independent sur face-matching algorithm gives a rigid body transforma tion, which allows the transfer of information between both modalities. A three-dimensional (3D) extension of the chamfer-matching principle makes up the core of this surface-matching algorithm. The optimal transformation is inferred from the minimization of a quadratic general ized distance between discrete surfaces, taking into ac count between-modality differences in the localization of A number of approaches to the analysis of phys iological data obtained from positron emission to mography (PET) require complementary anatomi cal information from another modality such as mag netic resonance imaging (MRI) (Mazziotta et aI., 1991). Since scans are not (and often cannot be) performed with perfectly reproducible patient posi tioning, increasing needs for accurate and reproduc ible three-dimensional (3D) registration methods have appeared. Most of the existing methods in volve user interaction. Procedural approaches, which rely on specific acquisition protocols [stereo taxic frames (Clarysse et aI., 1991), headholders (Bettinardi et aI., 1991), external markers (Koeppe et aI., 1991; Maguire et aI., 1991)], suffer from lack of versatility. Retrospective assisted approaches, 749 the segmented surfaces. The minimization process is ef ficiently performed via the precomputation of a 3D dis tance map. Validation studies using a dedicated brain shaped phantom have shown that the maximum registra tion error was of the order of the PET pixel size (2 mm) for the wide variety of tested configurations. The soft ware is routinely used today in a clinical context by the physiCians of the Service Hospitalier Frt!deric loliot (> 150 registrations performed). The entire registration process requires -5 min on a conventional workstation.

Multimodal neuroimaging provides a highly consistent picture of energy metabolism, validating 31 P MRS for measuring brain ATP synthesis

“…Since PET provided images of the whole brain, it was necessary to localize the 8-mL VOI on PET images to assess CMRglc from this volume only. We registered 3D reconstructed PET images with 3D MRI by using robust and fully automated rigid registration method (48). Registration allowed one to accurately localize the NMR detected VOI within PET images and to extract 18 F time activity from this VOI.…”

Simultaneous correction for interscan patient motion and geometric distortions in echoplanar imaging

“…The coregistration program is based on a surface matching algorithm (7) and involves two primary steps. In the first primary step, the brain surfaces from the two scans (high-resolution reference and EPI) are extracted semi-automatically using an extension of the techniques described by Alpert et al (9) and Mangin et al (10). First, pixels with a signal intensity outside that of the brain intensity range are excluded by applying an intensity window that is automatically calculated from the intensity histogram of the entire scan.…”

3-D chamfer distances and norms in anisotropic grids

“…These calculation have been extended to larger masks [23,24] and to higher dimensions [7]. Anisotropic lattices have also been considered [8,25,26]. However, those calculations remain tedious, are not systematized and thus have to be conducted manually for every mask size or anisotropy value.…”