Semiautomated segmentation of ovarian follicular ultrasound images using a knowledge-based algorithm

The application of a knowledge-based segmentation method to the problem of automatically detecting the outer follicle wall boundary in ultrasonographic images of ovaries is presented. A combination of computer detection and interactive adjustment was used to define an approximate inner follicle-wall boundary, which was then used by the computer algorithm as a priori knowledge to automatically find the outer follicle-wall border. The segmentation algorithm was tested on ultrasonographic images of women's ovaries that were imaged in vivo. The semiautomatic segmentations were compared to segmentations by an expert human observer in terms of border placement differences and in terms of quantitative parameters relevant to the physiologic status of the follicles. These physiological parameters include total and specific signal intensity from the follicle and from the follicle wall. The computer-detected outer follicle wall boundaries correlated well with the human observer-defined wall boundaries, in terms of enclosed follicle area, specific and total follicle signal, enclosed wall area, and specific and total wall signal. The actual border placement differences were also small, with a maximum placement difference of 1.47 +/- 0.83 mm and a root mean square (r.m.s.) placement difference of 0.59 +/- 0.28 mm.