Diagnose breast cancer through mammograms, using image processing techniques and optimization techniques

Microcalcifications are one of the key symptoms facilitating early detection of breast cancer. In this paper, The textural features are extracted from the segmented mammogram image to classify the microcalcifications into benign, malignant or normal. The reduced features are selected from the extracted set of features using reduction algorithms. Initially the reduced features are normalized between zero and one. The normalized feature values are given as input to a three-layer BPN to classify the microcalcifications into benign, malignant or normal. The network is trained to produce the output value 0.9 for malignant, 0.5 for benign and 0.1 for normal images. The BPN classifier is validated using ten fold validation Method. A Receiver Operating Characteristics (ROC) analysis is performed to evaluate the classification performances of the proposed approaches. The area under the ROC curve is used as a measure of the classification performance and it is denoted by Az. A larger value of Az indicates better classification performance. The proposed system is tested on 161 pairs of digitized mammograms from the MIAS database to establish its competence.