Advancements in Automated Diagnostic Mammography

This paper reports the results of an initial false positive (FP) investigation using a full set of mammogram risk factors, using a combination of logistic regression (LR), statistical analyses, and receiver operating characteristic (ROC) curves derived from LR density functions. These risk factors are important for this type of analysis and, until now, values have not been available in our data. Preliminary results provide evidence that combining information from different sources has the possibility of enhancing the model´s predictive capability. The non-image based features, in combination with a spatial feature, sometimes provide better performance than either set of feature(s) alone. These findings indicate that a more thorough investigation of the spatial attributes of conventional mammograms in combination with the full battery of breast cancer risk factors may be a useful path to follow in building models that are able to reduce the FP problem in mammography. Finally, this approach and paradigm may be applied to any dataset whose features are similarly characterized.