Authors : A.A. Kayode, N.O. Akande and E.O. Asani
Abstract: In the present day, the most rampant cancer discovered among women in various parts of the world is breast cancer. Early detection and diagnosis of breast cancer which can be achieved through mammography increases treatment options and a cure is more likely. In order to diagnose breast cancer, radiologists carefully examine patient’s X-ray images of the breast (mammograms) to see if there are significant visually extractable features that indicate the presence of breast cancer. However, visual features are subjective and diagnostic decisions should not be based on them because they are a function of radiologist’s opinion and experience. Thus, to eliminate the differential interpretations of abnormalities seen on mammograms among radiologists it is expedient to use computers to aid the extraction and selection of features which are not necessarily visually extractable. This study makes use of patient’s mammograms acquired from Radiology Department, Obafemi Awolowo University Teaching Hospital Complex Ile-Ife, Nigeria. Features are extracted from the mammograms using feature descriptors from Gray Level Co-occurrence Matrix (GLCM) and most discriminating features are selected using the proposed hybrid feature selection algorithm which is implemented to improve the classification accuracy. For each of the input image, the algorithm automatically selects relevant features from the set of extracted features. This algorithm reduces the extracted features by selecting the most relevant features thereby finding (near) optimal classification model of breast mammographic images. Two methods are combined for selecting optimal features viz.: the sequential forward selection and the Genetic algorithm. This is done, so as to cover the disadvantages of each one by the advantages of the other.
A.A. Kayode, N.O. Akande and E.O. Asani, 2019. Implementation of a Hybrid Feature Selection Algorithm for Improving Classification of Mammograms. Journal of Engineering and Applied Sciences, 14: 4419-4429.