Machine learning techniques and mammographic risk assessment

Breast tissue characteristics are widely accepted as important indicators of the likelihood of the developing breast cancer. Methods which have the ability to automatically classify breast tissue distribution therefore provide important tools in assessing the risk to which patients are exposed. This paper examines the machine learning techniques employed for knowledge discovery in a recent approach to mammographic risk assessment. A number of weaknesses for selected classification techniques are identified and examined. Additionally, important trends in the data such as decision class confusion and how this affects the ability to perform accurate knowledge discovery on the extracted image data are also explored. The paper is concluded with some ideas as to how the identified trends in the data and weaknesses in the classification approaches could be addressed.