QUANTITATIVE CLASSIFICATION OF BREAST-TUMORS IN DIGITIZED MAMMOGRAMS

The goal of this study was to develop a technique to distinguish benig n and malignant breast lesions in secondarily digitized mammograms. A set of 51 mammograms (two views/patient) containing lesions of known p athology were evaluated using six different morphological descriptors: circularity, mu(R)/sigma(R) (where mu(R)=mean radial distance of tumo r boundary, sigma(R)=standard deviation); compactness, P-2/A (where P= perimeter length of tumor boundary and A=area of the tumor); normalize d moment classifier; fractal dimension; and a tumor boundary roughness (TBR) measurement (the number of angles in;the tumor boundary with mo re than one boundary point divided by the total number of angles in th e boundary). The lesion was segmented from the surrounding background using an adaptive region growing technique. Ninety-seven percent of th e lesions were segmented using this approach. An ROC analysis was perf ormed for each parameter and the results of this analysis were compare d to each other and to those obtained from a subjective review by two board-certified radiologists who specialize in mammography. The result s of the analysis indicate that all six parameters are diagnostic for malignancy with areas under their ROC curves ranging from 0.759 to 0.9 28. We observed a trend towards increased specificity at low false-neg ative rates (0.01 and 0.001) with the TBR measurement. Additionally, t he diagnostic accuracy of a classification model based on this paramet er was similar to that of the subjective reviewers. (C) 1996 American Association of Physicists in Medicine.