Authors : S. Sasikala and M. Ezhilarasi
Abstract: Computer Aided techniques developed for diagnosing the breast cancer plays a vital role in the early diagnosis of breast cancer and treatment in reducing the mortality rate. Better accuracy will generally be achieved using a combination of features instead of single type of feature descriptor. This research aims to improve the diagnostic accuracy and to reduce the false positive detection. Six different descriptors and their combination have been used to represent mammographic texture. The individual and combined feature vectors are reduced by principal component analysis and then classified by a multilayer Perceptron neural network using back propagation algorithm. The performance of the classification is evaluated with the texturefeaturesseparately and theircombination ie. the concatenation of the feature vectors from individual feature extraction techniques on the Digital Database for Screening Mammography (DDSM) and INbreast database by computing various performance metrics. The results show that the use of feature combination improves the performance of classification when a system cannot be tuned to an individual dataset. Eighteen performance metrics including Accuracy, Sensitivity, Specificity, Mathews Correlation coefficient, F1 score, discriminant power, Youden’s index etc. Al these metrics were improved for the combined features for both dataset.
S. Sasikala and M. Ezhilarasi, 2016. Combination of Mammographic Texture Feature Descriptors for Improved Breast Cancer Diagnosis. Asian Journal of Information Technology, 15: 4054-4062.