Radial-searching contour extraction method based on a modified active contour model for mammographic masses

Abstract

In this study, we developed an automatic extraction scheme for the precise recognition of the contours of masses on digital mammograms in order to improve a computer-aided diagnosis (CAD) system. We propose a radial-searching contour extraction method based on a modified active contour model (ACM). In this technique, after determining the central point of a mass by searching for the direction of the density gradient, we arranged an initial contour at the central point, and the movement of a control point was limited to directions radiating from the central point. Moreover, it became possible to increase the extraction accuracy by sorting out the pixel used for processing and using two images-an edge-intensity image and a degree-of-separation image defined based on the pixel-value histogram-for calculation of the image forces used for constraints on deformation of the ACM. We investigated the accuracy of the automated extraction method by using 53 masses with several "difficult contours" on 53 digitized mammograms. The extraction results were compared quantitatively with the "correct segmentation" represented by an experienced physician's sketches. The numbers of cases in which the extracted region corresponded to the correct region with overlap ratios of more than 81 and 61% were 30 and 45, respectively. The initial results obtained with this technique show that it will be useful for the segmentation of masses in CAD schemes.

The power laws: Zipf and inverse Zipf for automated segmentation and classification of masses within mammograms

“…Once the contour is initialized, it moves according to the local characteristics. Among methodologies existing in the literature belonging to edge-based technique, we can cite the work proposed in (Nakagawa et al 2008), where authors developed an automatic masses-contour extraction scheme to recognize the unclear contour of a mass within mammograms; it consists of a radial-searching contour extraction approach based on a modified active contour model: radial searching contour extraction method (ACM). Once the central point of a mass is determined through the search of the density gradient direction, authors perform the initialization of the contour at the central point and limit the movement of a control point to directions radiating from the central point.…”

Snakules: A Model-Based Active Contour Algorithm for the Annotation of Spicules on Mammography

“…While the algorithm developed by Qian et al is able to detect spicules of different sizes, it suffers from some drawbacks, most notably translation/rotational dependence of the decomposition used and ad hoc spicule aggregation rules with lack of model-based design mechanisms. There are other groups that have attempted to deploy snake-like devices to segment solid masses and use feature extraction strategies to classify the masses (e.g., [19]- [21]); however, to the best of our knowledge, our approach is the first attempt aimed at explicitly capturing the path of spicules using deformable active contours in a bid to use them for distinguishing spiculated masses from other breast structures.…”