Catálogo de publicaciones - libros

This paper presents an algorithm for classifying different tissue types in T1-weighted MR brain images using fuzzy segmentation. The main aim in this study is to compensate for the blurring effect on tissue boundaries due to partial volume effects. This paper is organized as follows: first, an adaptive greedy contour model has been developed to separate the intracranial volume (ICV) from the scalp and skull. Second, in order to deal with the problem of the partial volume effect, an algorithm for fuzzy segmentation is presented which has integrated fuzzy spatial affinity with statistical distributions of image intensities for each of the three tissues – cerebrospinal fluid, white matter and grey matter. This algorithm is tested on well-established simulated MR brain volumes to generate an extensive quantitative comparison with different noise levels and different slice thicknesses ranging from 1mm to 5mm. Finally, the results of this algorithm on clinical MR brain images are demonstrated.

With the emergence of swept-volume ultrasound (US) probes, precise and almost real-time US volume imaging has become available. This offers many new opportunities for computer guided diagnosis and therapy, 3-D images containing significantly more information than 2-D slices. However, computer guidance often requires knowledge about the exact position of US voxels relative to a tracking reference, which can only be achieved through probe calibration. In this paper we present a 3-D US probe calibration system based on a membrane phantom. The calibration matrix is retrieved by detection of a membrane plane in a dozen of US acquisitions of the phantom. Plane detection is robustly performed with the 2-D Hough transformation. The feature extraction process is fully automated, calibration requires about 20 minutes and the calibration system can be used in a clinical context. The precision of the system was evaluated to a root mean square (RMS) distance error of 1.15mm and to an RMS angular error of 0.61°. The point reconstruction accuracy was evaluated to 0.9mm and the angular reconstruction accuracy to 1.79°.