An automated mammogram classification system using modified support vector machine

Abstract

Aderonke Anthonia Kayode,1 Noah Oluwatobi Akande,1 Adekanmi Adeyinka Adegun,1 Marion Olubunmi Adebiyi21Computer Science Department, Landmark University, Omu-Aran, Kwara State, Nigeria; 2Department of Computer and Information Sciences, Covenant University, Ota, Ogun State, NigeriaPurpose: Breast cancer remains a serious public health problem that results in the loss of lives among women. However, early detection of its signs increases treatment options and the likelihood of cure. Although mammography has been established to be a proven technique of examining symptoms of cancer in mammograms, the manual observation by radiologists is demanding and often prone to diagnostic errors. Therefore, computer aided diagnosis (CADx) systems could be a viable alternative that could facilitate and ease cancer diagnosis process; hence this study.Methodology: The inputs to the proposed model are raw mammograms downloaded from the Mammographic Image Analysis Society database. Prior to the classification, the raw mammograms were preprocessed. Then, gray level co-occurrence matrix was used to extract fifteen textural features from the mammograms at four different angular directions: θ={0°, 45°, 90°, 135°}, and two distances: D={1,2}. Afterwards, a two-stage support vector machine was used to classify the mammograms as normal, benign and malignant.Results: All of the 37 normal images used as test data were classified as normal (no false positive) and all 41 abnormal images were correctly classified to be abnormal (no false negative), meaning that the sensitivity and specificity of the model in detecting abnormality is 100%. After the detection of abnormality, the system further classified the abnormality on the mammograms to be either “benign” or “malignant”. Out of 23 benign images, 21 were truly classified as benign. Also, out of 18 malignant images, 17 were truly classified to be malignant. From these findings, the sensitivity, specificity, positive predictive value, and negative predictive value of the system are 94.4%, 91.3%, 89.5%, and 95.5%, respectively.Conclusion: This article has further affirmed the prowess of automated CADx systems as a viable tool that could facilitate breast cancer diagnosis by radiologists.Keywords: cancer diagnosis, CADx systems, radiologists, diagnostic errors, GLCM

Keywords

• Cancer diagnosis
• CAD systems
• Radiologists
• Diagnostic errors
• Gray level co-occurrence matrix