Data Mining Framework for Breast Cancer Detection in Mammograms: A Hybrid Feature Extraction Paradigm

摘要

Normally, it will start when a breast cell becomes abnormal and this abnormality can either be benign or malignant. In Singapore, an average of 1490 woman (29%) were diagnosed with breast cancer and the number of cases has been steadily increasing from 2005 to 2009. Breast cancers are on a trajectory to become the greatest proportion of cancers deaths. The probability of men contracting breast cancer is about 100 times lower than the risk for women, but the survival rate of breast cancer for men is about the same as for women. This survival rate depends on in which stage the cancer was detected. In general, the rates get better the earlier the cancer is detected. Therefore, an early diagnosis is very important to increase the live span of a patient. Objectives: To develop a Computer-Aided Diagnosis (CAD) system to detect cancer cells from mammogram images. Perform data mining on these images to characterize and classify normal, benign and malignant breast tumors. Methods: Prior to feature extraction, the mammogram images were standardized through pre-processing. The feature extraction itself was carried out using the Discrete Wavelet Transform (DWT). After that, we identified normal, benign and malignant breast tumors with the help of different classifiers. Results: The results of the feature extraction step were subjected to the Analysis of Variance (ANOVA) test, which assessed their significance. Only clinically significant features were used for classification. From all the classifier we have analyzed, the Fuzzy Sugeno Classifier (FSC) provided the best result. It can diagnose the three classes with 82.9% of accuracy, 94.9% of Positive Predictive Value (PPV), 96.7% for sensitivity and 94.4% for specificity. Conclusion: Mammography based CAD systems can aid practitioners during their task of breast cancer diagnosis and treatment monitoring.