Segmenting lesions is a vital step in many computerized mass-detection
schemes for digital (or digitized) mammograms, We have developed two
novel lesion segmentation techniques-one based on a single feature cal
led the radial gradient index (RGI) and one based an simple probabilis
tic models to segment mass lesions, or other similar nodular structure
s, from surrounding background, In both methods a series of image part
itions is created using gray-level information as well as prior knowle
dge of the shape of typical mass lesions, With the former method the p
artition that maximizes the RGI is selected. In the latter method, pro
bability distributions for gray-levels inside and outside the partitio
ns are estimated, and subsequently used to determine the probability t
hat the image occurred for each given partition, The partition that ma
ximizes this probability is selected as the final lesion partition (co
ntour), We tested these methods against a conventional region growing
algorithm using a database of biopsy-proven, malignant lesions and fou
nd that the new lesion segmentation algorithms more closely match radi
ologists' outlines of these lesions, At an overlap threshold of 0.30,
gray level region growing correctly delineates 62% of the lesions in o
ur database while the RGI and probabilistic segmentation algorithms co
rrectly segment 92% and 96% of the lesions, respectively.