Breast lesion classification by computer and xeroradiograph

One hundred and twenty pathologically proven lesions were digitized by computer and recorded on film. Four measures of malignancy, calcification, spiculation, roughness, and area‐to‐perimeter ratio then were mathematically extracted from the digitized xeroradiograph lesions. Using three methods of classification, these were identified by the computer as malignant and benign. Two radiologists then classified the photographed lesions. Comparison of the computer classification and the radiologists' classification using operational characteristic curves showed the computer to perform as well as an expert radiologist.