IMAGE-ANALYSIS AND MACHINE LEARNING APPLIED TO BREAST-CANCER DIAGNOSIS AND PROGNOSIS

Fine needle aspiration (FNA) accuracy is limited by, among other facto rs, the subjective interpretation of the aspirate. We have increased b reast FNA accuracy by coupling digital image analysis methods with mac hine learning techniques. Additionally, our mathematical approach capt ures nuclear features (''grade'') that are prognostically more accurat e than are estimates based on tumor size and lymph node status. An int eractive computer system evaluates, diagnoses and determines prognosis based on nuclear features derived directly from a digital scan of FNA slides. A consecutive series of 569 patients provided the data for th e diagnostic study. A 166-patient subset provided the data for the pro gnostic study. An additional 75 consecutive, new patients provided sam ples to test the diagnostic system. The projected prospective accuracy of the diagnostic system was estimated to be 97% by 10-fold cross-val idation, and the actual accuracy on 75 new samples runs 100%. The proj ected prospective accuracy of the prognostic system was estimated to b e 86% by leave-one-out testing.