AUTOMATIC SEGMENTATION AND CLASSIFICATION OF PHANTOM IMAGES FOR MAMMOGRAPHY QUALITY CONTROL

Abstract

Breast cancer is one of the most common forms of cancer in women worldwide. At present, mammography is one of the most reliable methods for early breast cancer detection. However, mammography images are difficult to interpret by the experts owing to the fact their features are typically very small, have poor contrast and a wide range of anatomical patterns. In order to evaluate the quality of the mammographic facilities, the visibility and the scoring of features in phantom image is used. The goal of this research was to automate breast phantom image scoring using image processing techniques. First breast phantom films were digitized. For each category of structures, sub-images were extracted from the digitized phantom. Artifacts were removed using extraction process and noise was removed by the 2D median filter. Contrast-limited adaptive histogram equalization (CLAHE) algorithm was used to improve the appearance of the image. The morphological features of the phantom images were also calculated. Ten digitized phantom images were extracted into sub-images and 150 sample sub images were considered for the evaluation. Artificial Neural Network (ANN) was used as classifiers. For each classifier, the performance factor such as sensitivity, specificity and accuracy are computed. All the techniques were implemented through MATLAB R2013a platform. It is observed that the proposed scheme with ANN classifier out performed by giving 96% accuracy, 95% sensitivity, 92.5% specificity and 4% the probability of misclassification error to classify the phantom images as fiber, specks or mass