search railabs.duhs.duke.edu.

Papers Published

1. Floyd, C.E., Jr. and Lo, J.Y. and Baker, J.A., Prediction of breast biopsy outcomes from mammographic findings, Computer-Aided Diagnosis in Medical Imaging. Proceedings of the First International Workshop on Computer-Aided Diagnosis (1999), pp. 193 - 200 .
   (last updated on 2007/04/15)

   Abstract:
   Describes a computer aid to predict the malignancy of nonpalpable lesions that are examined with diagnostic mammography and are considered for biopsy. The goal is to improve the specificity of diagnosis with little loss of sensitivity, thus significantly improving the positive predictive value of breast biopsy. An artificial neural network (ANN) is described to assist radiologists in the differentiation of benign from malignant lesions. Inputs to the ANN were derived from the patient's history and the radiologist's description of lesion morphology following the ACR Breast Imaging Reporting And Data System (BI-RADS^TM). The output of the neural network is the likelihood of malignancy. Evaluation of the system on 500 cases demonstrates that 22% of the benign biopsies could be avoided without missing a malignancy. At this threshold, the positive predictive value (PPV) of biopsy would be improved from 35% to 41%. With a less conservative approach, 41% of the benign biopsies could be avoided while still performing biopsies on 98% of the malignancies. At this threshold, the PPV of biopsy would be improved from 35% to 47%

   Keywords:
   cancer;mammography;medical diagnostic computing;medical image processing;neural nets;surgery;