Visual versus Statistical Features Selection Applied to Mammography Mass Detection
Ibrahim Mohamed Ibrahim, Manal Abdel Wahed · Journal of Medical Imaging and Health Informatics · 2014
Breast cancer is the highest frequent form of cancer in women today. Mammography is the most reliable and practical method capable of detecting breast cancer at its early stage. Physicians ’ ability to detect tumors can be assisted by using some computerized features extraction algorithms. The key point is to use the most significant features with the most suitable classifiers. This paper presents a comparative study between two statistical feature selection methods and an immature, newly used visual method that comes from the bioinformatics field. The visual method puts the values of the extracted features in a heat map and then assigns different colors for these values. After that, it searches for some visual patterns in these colors that show discrimination power between the different breast tissue types. 61 features were extracted to differentiate between normal and tumorous breast tissues. The discriminatory power of these features was tested using the visual method in a comparison with other two statistical methods: the t-test and the Kullback–Leibler (KL) divergence method. For classification, we used the minimum distance, the k-Nearest Neighbor (k-NN), and the Support Vector Machine (SVM) classifiers. Results show that the visual method needs some pre-processing steps, to overcome noise and database outliers, to provide higher performance over the other two statistical methods.