Screening for breast cancer involves the examination of large numbers of mammograms, which involves a large proportion of the time of the staff available and adds to delays in the early diagnosis of breast cancer. Automatic analysis of mammograms to exclude those without any indication of cancer, if feasible, would allow skilled effort to be concentrated on mammograms with indications of cancer. Lesions on the surface of the mammogram are easily detected, but internal occult lesions may only be apparent by the change in texture in the area of the growth. An example of a malignant lesion is given, taken from the MIAS database and contour plots of graylevel are included. It can be seen from the contour plots that the boundary of the lesion is not smooth or well-defined. A graph of a typical cross-section of the edge of the lesion shows that the graylevels increase from values around 150 outside the lesion to a plateau around 200 in the lesion, with a steep, but variable, gradient in between. It is apparent that the texture between the 150 and 200 graylevel contours is affected by the considerable gradient in the graylevel. The paper proposes a method of investigating the texture in the context of significant graylevel gradients.
展开▼