Computer-Aided Diagnosis of Mammographic Masses

摘要

A new Model-Based Vision algorithm was developed to find possibly cancerousregions of interest (ROIs) in digitized mammograms and to correctly identify the malignant masses. This work has shown a sensitivity of 92 percent for locating malignant ROIs. The database contained 272 images (12 bit, 1OO microns) with 36 malignant and 53 benign mass images. Of the 53 biopsied benign cases, 74 percent were correctly classified. The Focus of Attention (segmentation) Module algorithm used a physiologically motivated Difference of Gaussians (DoG) filter to highlight mass-like regions in the mammogram. The Index Module labeled the regions by their hypothesized class: large or medium mass. Then it used size, shape, and contrast tests to reduce the number of non-malignant regions from 8.4 to 2.8 per image. Size, shape, contrast, and Laws texture features were used to develop the Prediction Module's mass model. Statistical and derivative-based feature saliency techniques were used to determine the best features. Nine features were chosen to define the model. Using this model, the Matching Module classified the regions using a multilayer perceptron neural network architecture trained with an imbalanced training set weight update algorithm to obtain an overall classification accuracy of 100 percent for the segmented malignant masses with a false-positive rate of 1.8/image.