Breast Cancer Diagnosis Using Data Mining Methods, Cumulative Histogram Features, and Gary Level Co-occurrence Matrix

Background: Breast cancer is the most prevalent cancer and the second leading cause of cancer death among women. Recently, computerized methods have proved to be an extraordinary tool for radiologists for diagnosing breast cancer. In recent years, data mining techniques and image mining of digital mammography have played a significant role in diagnosing breast cancer. Mammography is considered to be an efficient tool for the early diagnosis of breast cancer. Methods: Mammography images are classified into three groups of benign, malignant, and normal images. This paper proposes a method for optomizing breast cancer diagnosis in digital mammography using Gray Level Co-occurrence Matrix (GLCM) and cumulative histogram features. In the proposed method, a combination of Imperialist Competitive Algorithm (ICA) and decision tree method was introduced for classifying mammogram images by means of GLCM and cumulative histogram features. The proposed approach is referred to as DC-MICs algorithm. Results: In this study, midified imperialist competitive algorithm was implemented for selecting and reducting useless features. Decision tree method was aimed at classifying images based on extracted features. Conclusion: The efficiency of this method was investigated on Digital Database for Screening Mammography (DDSM) breast cancer through the accuracy criterion and the selection of appropriate features for classification. The obtained results indicated that the proposed DC-MICs algorithm has remarkably improved the accuracy, sensitivity, specificity and F-score measures