J Braz Comput Soc (2011) 17:117–132 DOI 10.1007/s13173-011-0032-8 ORIGINAL PAPER Detection of point landmarks in 3D medical images via phase congruency model Ricardo J. Ferrari · Stéphane Allaire · Andrew Hope · John Kim · David Jaffray · Vladimir Pekar Received: 16 September 2010 / Accepted: 28 February 2011 / Published online: 19 March 2011 © The Brazilian Computer Society 2011 Abstract This paper presents a novel technique for detec- tion of point landmarks in volumetric medical images based on a three-dimensional (3D) Phase Congruency (PC) model. A bank of 3D log-Gabor filters is specially designed in the frequency domain and used to compute 3D energy maps, which are further combined to form the phase congruency measure. The PC measure is invariant to intensity variations and contrast resolution and provides a good indication of feature significance in an image. To detect significant 3D point landmarks, eigen-analysis of a 3 × 3 matrix of second- order PC moments, computed for each point in the image, is performed followed by local maxima detection. Two differ- ent application scenarios in radiation therapy planning of the head and neck anatomy are used to illustrate the feasibility and usefulness of the proposed method. Keywords Point landmarks · 3D hase congruency · 3D log-Gabor filters · Wavelets · Nonrigid registration · Radiation therapy R.J. Ferrari ( ) Department of Computer Sciences, Federal University of São Carlos (UFSCar), Rod. Washington Luís, Km 235, Caixa Postal 676, 13565-905 São Carlos, SP, Brazil e-mail: rferrari@dc.ufscar.br S. Allaire · A. Hope · J. Kim · D. Jaffray Princess Margaret Hospital, University of Toronto, Toronto, ON, Canada V. Pekar Philips Research North America, Toronto, Markham, ON, Canada 1 Introduction Detection of distinctive 3D anatomical point landmarks is an important task in a variety of medical image analysis applications including image-to-image and atlas-to-image registration. Landmark-based registration usually comprises three main steps: (1) detection of reliable landmarks in datasets to be registered; (2) establishment of correspon- dences between the landmarks; and (3) determination of the transformation between the datasets using the landmark cor- respondences. Usually, the localization of anatomical point landmarks in medical images is carried out manually by an expert physician. However, manual annotation of image landmarks, especially in 3D images, is a time-consuming and error-prone task, and in general it presents high inter- observer variability and low reproducibility [17]. Therefore, there is a great interest in developing automatic techniques that could reliably detect 3D landmarks in medical image. In this work, a new technique for detection of 3D point landmarks in medical images based on an implementation of the 3D PC model is proposed. The intrinsic image contrast invariance as well as the good spatial and frequency local- ization provided by the log-Gabor filters make the proposed method very suitable for applications in medical image anal- ysis. In addition to the landmarks, the method also provides 3D PC maps, which correspond to important visual salient features in an image and can be used, for instance, to re- place image intensity information in image registration ap- plications, e.g., multimodality registration [46]. The paper is organized as follows: Section 2 describes the related previous work presented in the literature; the methodology of the proposed method is described in Sect. 3, followed by two different application scenarios in Sect. 4 that are used to illustrate the feasibility and usefulness of the 3D PC-landmarks algorithm. Conclusions are presented in Sect. 5.