Artificial neural network: improving the quality of breast biopsy recommendations.

PURPOSE: To evaluate the performance and inter- and intraobserver variability of an artificial neural network (ANN) for predicting breast biopsy outcome. MATERIALS AND METHODS: Five radiologists described 60 mammographically detected lesions with the American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) nomenclature. A previously programmed ANN used the BI-RADS descriptors and patient histories to predict biopsy results. ANN predictive performance was compared with the clinical decision to perform biopsy. Inter- and intraobserver variability of radiologists' interpretations and ANN predictions were evaluated with Cohen kappa analysis. RESULTS: The ANN maintained 100% sensitivity (23 of 23 cancers) while improving the positive predictive value of biopsy results from 38% (23 of 60 lesions) to between 58% (23 of 40 lesions) and 66% (23 of 35 lesions; P < .001). Interobserver variability for interpretation of the lesions was significantly reduced by the ANN (P < .001); there was no statistically significant effect on nearly perfect intraobserver reproducibility. CONCLUSION: Use of an ANN with radiologists' descriptions of abnormal findings may improve interpretation of mammographic abnormalities.

Duke Scholars

Author Joseph Yuan-Chieh Lo Radiology

Author Jay Alan Baker Radiology, Breast Imaging