Extraction and Measurement of Lesion Edge Blur in Mammograms by Computer Image Analysis

The overall objective of the project is to derive quantitative descriptions of diagnostically important features of lesions in mammograms and thus provide radiologists with an additional diagnostic tool. The work presented here is concerned with the extraction of edge blur measures in circumscribed lesions by computer image analysis. Edge definition is an important diagnostic sign: a blurred contour indicates probable malignancy. Background detail, which interferes with the lesion’s edge information, was removed without significantly affecting edge blur. Different approaches were needed for removing large structures, characterised by low-frequency, and for small structures, characterised by high frequency. After lesion extraction, the blur measures derived for the lesions could be differentiated with greater precision than could be achieved visually. The measures will be correlated with histological data to determine their significance for diagnosis.