The idea of fitting warped ellipses to the TRUS images and a final warped ellipsoid to the resulting contours in a report by Badiei et al. (14) is extended to fitting tapered and warped ellipses and a tapered and warped ellipsoid to obtain

better fitting contours. The posterior warping is required to account for the posterior deformation of the gland caused by the presence of the ultrasound probe and the tapering parameter is added for better agreement with the anatomy of the prostate. Because fitting such 2D and 3D shapes to the TRUS images may be computationally expensive, the TRUS images themselves are deformed to result in elliptical cross-sections of the gland. Fitting an ellipse is a fast and straightforward problem. Figure 1 BI 2536 molecular weight shows the main steps of the semiautomatic segmentation algorithm. The algorithm is initiated by the user identifying the base, apex, and midgland images; the TRUS probe center; and six boundary points on the midgland image. The base and apex images are images in which the most superior and inferior portions of the prostate are visible. The six boundary points include:

p1 = lowest lateral; p2 = lateral right; p3 = midposterior; p4 = midanterior; and two points, p5 and p6, guided by points p1, p2, and Trichostatin A solubility dmso p4. These points are selected to extract the size, amount of warping, and the transverse tapering of the prostate boundary while eliminating the variability of point selection by directing the user to specific regions (Fig. 1a). By knowing the location of the TRUS center and the lowest lateral and midposterior points, all TRUS images are unwarped to remove the posterior deformation. A tapered ellipse Ribonucleotide reductase is then fitted

to the initial points and their reflections with respect to the medial line. The resulting tapering value, 0≤t≤10≤t≤1 (t=0t=0 being an ellipse), is used to untaper the TRUS images in the transverse plane. It is assumed that tapering linearly reduces to zero toward the base and apex, with these two regions having elliptical cross-sections. The midgland tapering value is also used to untaper the initial tapered ellipse contour in the midgland slice to obtain an initial elliptical contour on this slice. The interacting multiple model probabilistic data association (IMMPDA) edge detection algorithm introduced by Abolmaesumi and Sirouspour (19) is then used to search for the boundary of the prostate within a neighborhood of less than 0.5 cm inside and outside the initial midgland ellipse (Fig. 1b). In effect, the IMMPDA algorithm acts to leverage a coarse set of manually selected points to guide a higher resolution detection of the prostate boundary using statistical sampling techniques designed to suppress the type of image noise typically found in ultrasound images.