Discovering Potential Precursors of Mammography Abnormalities Based on Textual Features, Frequencies, and Sequences

Diagnosing breast cancer from mammography reports is heavily dependant on the time sequences of the patient visits. In the work described, we take a longitudinal view of the text of a patient’s mammogram reports to explore the existence of certain phrase patterns that indicate future abnormalities may exist for the patient. Our approach uses various text analysis techniques combined with Haar wavelets for the discovery and analysis of such precursor phrase patterns. We believe the results show significant promise for the early detection of breast cancer and other breast abnormalities.