Performance of Support Vector Machine (SVM) using Binary Robust Invariant Scalable Keypoints (BRISK) with different RBF kernel scales in classifying types of breast tissue

Breast cancer is the major cause of death among women. The detection of breast cancer tissue is important for radiologists to diagnose breast cancer. Computer Aided Diagnosis (CADx) is one of the diagnosis systems to assist radiologists with an accurate diagnosis as well as to reduce the error rate. Hence, the objectives of the study are to propose a CADx for the classification of breast tissue. A set of 322 data was obtained from mini–Mammogram Image Analysis Society (MIAS) database used in the implementation of the CADx. The data is pre-processed first by using median filter and histogram stretching methods before segmented using hybrid Otsu and Mathematical Morphology. Then the features are detected and extracted by using the Binary Robust Invariant Scalable Keypoints (BRISK) features detectors. Next, the detected features are selected by using Principal Component Analysis (PCA) to reduce the dimensionality issue. Finally, the Support vector machine (SVM) is applied to classify the breast cancer tissue. In the classification stage, different values of Radial Basis Function (RBF) kernel scales are experimented concurrently with the box constraint levels. The accuracy of the proposed CADx is evaluated to measure its performance. The results demonstrate an accuracy of 83% is obtained which concludes that the proposed CADx has impressive performance.