Mammographic risk assessment: does anatomical linear structure information improve the accuracy of a density-based classifier?

Mammographic risk assessment is concerned with the probability of a woman developing breast cancer. Recently, it has been suggested that the density of linear structures is related to risk. Two independent sets of mammographic images were annotated according to BIRADS risk classes by expert radiologists. Linear structure information was extracted from each image using the line operator method, and density segmentation was performed using a method based on minimum error thresholding.

Linear discriminant analysis and a Support Vector Machine classifer were used to classify the images in to BIRADS classes. The classification was performed three times for each dataset – once using density information only, once using linear structure information only, and once using both density and linear structure information. The results of classification showed a marked improvement when both density and linear structure information were used, suggesting that linear structure information is valuable in mammographic risk classification