Computer-aided diagnosis of breast cancer via Gabor wavelet bank and binary-class SVM in mammographic images

• Published in: Journal of experimental & theoretical artificial intelligence Vol. 28; no. 1-2; pp. 295 - 311
• Main Authors: Torrents-Barrena, Jordina, Puig, Domenec, Melendez, Jaime, Valls, Aida
• Format: Journal Article
• Language: English
• Published: Abingdon Taylor & Francis 03.03.2016; Taylor & Francis Ltd
• Subjects: Artificial intelligence; Breast; Breast cancer; Cancer; Classification; Diagnosis; Diseases; Gabor filters; kernels; kurtosis; Mammography; Mathematical analysis; mean; Mortality; skewness; Standard deviation; Support vector machines; SVM; types of breast cancer; X-ray images
• ISSN: 0952-813X; 1362-3079
• DOI: 10.1080/0952813X.2015.1024491

Breast cancer is one of the most dangerous diseases that attack women in their 40s worldwide. Due to this fact, it is estimated that one in eight women will develop a malignant carcinoma during their life. In addition, the carelessness of performing regular screenings is an important reason for the increase of mortality. However, computer-aided diagnosis systems attempt to enhance the quality of mammograms as well as the detection of early signs related to the disease. In this paper we propose a bank of Gabor filters to calculate the mean, standard deviation, skewness and kurtosis features by four-sized evaluation windows. Therefore, an active strategy is used to select the most relevant pixels. Finally, a supervised classification stage using two-class support vector machines is utilised through an accurate estimation of kernel parameters. In order to show the development of our methodology based on mammographic image analysis, two main experiments are fulfilled: abnormal/normal breast tissue classification and the ability to detect the different breast cancer types. Moreover, the public screen-film mini-MIAS database is compared with a digitised breast cancer database to evaluate the method robustness. The area under the receiver operating characteristic curve is used to measure the performance of the method. Furthermore, both confusion matrix and accuracy are calculated to assess the results of the proposed algorithm.