Search

The spatial distribution of the optical density of an image of a cell nucleus can be imagined to resemble a rugged surface in a three-dimensional space. The area of this surface can be measured using different unit-sized squares, yielding different values for the total area depending on the unit size. A fractal dimension can be calculated from the relationship between the unit size and the overall area measured; this fractal dimension is characteristic of the surface measured. This technique was applied to extract three new nuclear features from stained cervical cells. These new features were useful for the computer discrimination of normal and premalignant cervical cells. The features are defined, and their discriminating power is compared to that of some other nuclear features.

An automated procedure that refines the nuclear contour of a previously segmented nucleus is described. The algorithm makes use of intensity information, edge magnitude information and both object and edge connectivity information. This automated procedure generates a closed contour precisely along the edge of the nucleus. The procedure was tested on a database of 3,680 red-green-blue images of thionin-SO2 and orange II-stained cervical cells obtained from normal and dysplastic samples. When used in conjunction with a simple threshold selection algorithm and an artifact removal routine, this edge relocation algorithm resulted in the correct segmentation of over 98% of the nuclei. Only 63 (1.7%) of all nuclei were incorrectly segmented.

The segmentation accuracy of four fast and simple gray-scale threshold selection methods were compared using a data base of stained cervical cell images. Some postprocessing was applied to the segmented images to increase the accuracy of the nuclear segmentation. The most accurate method correctly segmented the cytoplasm of 81% of the cell images and the nuclei of 78% of the cell images in the data base.

Multispectral images of stained cells enable the use of color differences to segment and/or to discriminate between image components, such as cell types and cellular subcomponents. When the spectral characteristics of the image components do not change over the area of a slide or from slide to slide, one can create a constant weighted linear combination of spectral images to generate one-dimensional or two-dimensional images that have the desired contrast between the image components that must be discriminated. However, when the spectral characteristics are not constant, i.e., when they vary from image to image, a constant weighted linear combination cannot be employed; instead, an appropriate solution must be found for each selected image. This is usually a time-consuming, manual procedure that cannot be employed in a fully automated process of discriminating and segmenting stained cells. This paper describes an algorithm that uses principal components decomposition basis vectors to generate a nonstatic weighted linear combination of color images that can be used by an automated system. This algorithm relies on a semiconstant relationship between the areas (sizes) of the image components that are to be discriminated and/or segmented. The technique has been successfully applied as an aid in the segmentation of images of stained cervical smears; the images were acquired with a three-chip CCD camera that generates three broad-band color images.