Computerized Diagnosis of Breast Fine-Needle Aspirates
William H. Wolberg, W. Nick Street, Olvi L. Mangasarian · The Breast Journal · 1997
Abstract: The goal of this work is to determine the accuracy of computer-based image analysis in diagnosing breast fine-needle aspirates (FNA). On 192 FNAs, the computer-based diagnostic accuracy was 97.9%. This is consistent with the 97.5% accuracy projected by machine learning methods during the initial training in 1994 with 569 FNAs. One of the attributes of this system is the rendering of a value that estimates the probability of malignancy. We consider estimated probability of malignancy values between 0.30 and 0.70 to be equivocal. Eleven of our samples (5.7%) fell into this equivocal category. All computer misclassified FNAs were correctly diagnosed visually based on contextual features. We propose this computer-based system as a diagnostic adjunct rather than as a stand-alone system.