ELSEVIER Pattern RecognitionLetters 16 (1995) 1033-1042 Pattern Recognition Le~ers Point landmarks for registration of CT and MR images S. Banerjee, D.P. Mukherjee, D. Dutta Majumdar* Electronics and Communication Sciences Unit, Indian Statistical Institute, Calcutta 700035, India Received 1 November1994; revised 17 March 1995 Abstract We propose here a point landmark based registration method utilizing geometric invariance properties of biomedical images. These point landmarks constitute entrance and exit points of concavities of individual structures and points of inflexion of curves, derived from the convex hull. Registration is performed in a canonical frame of reference. This technique is fast, semi-automatic and computationally inexpensive. Keywords: Registration; point Landmarks;Concavities;Affine and projectivetransformations;Canonical frame; Measure of mismatch 1. Introduction Registration is the process of determining a point- to-point correspondence between two images of the same object. This is a crucial first step for fusion of two image data sets to obtain an integrated image display. Applications include multimodality medical imaging (Banerjee and Dutta Majumder, 1993) and multisen- sor data fusion in remote sensing. Since images used in these applications are generally not well defined, the search for computationally inexpensive robust reg- istration methods which require little or no expert in- teraction, constitutes an open problem in image pro- cessing. In this paper we present our endeavors to develop a semi-automatic registration technique for 2D images which has been applied to register identical cross- sections of the human brain obtained from x-ray Com- puted Tomography (CT) and Magnetic Resonance (MR) modalities. Our method identifies point land- marks on surfaces of anatomical regions of interest * Corresponding author. (ROI) that are visible in all scans considered. These ROIs have a purely geometrical representation and can be extracted from the image by segmentation. The se- lection of such point landmarks is based on the geo- metrical invariance properties of the ROIs within the original image. Biomedical images possess a large number of concavities and we have used these con- cavities to locate landmark points or signatures. This method can be classified (Van den Elsen et al., 1993) as a direct, semi-automatic method using a lo- cal or global transformation, depending on the number of structures of interest in the image and can be de- scribed as interpolating or approximating depending on whether three or more points are used. As we shall elaborate later, this point landmark based method pos- sesses two distinct advantages over other point land- mark based methods. The use of landmarks in im- ages with geometric invariance properties simplifies the identification as well as automate the registration process. This method thus obviates the necessity of expert interaction (Hill et al., 1991 ) and hence results in a speed-up of computation. Secondly, this method does not require the computation of first and higher 0167-8655/95/$09.50 (~) 1995 Elsevier Science B.V. All rights reserved SSDI0167-8655(95)00058-5