Combining Feature Methods for Content-Based Classification of Mammogram Images
Keith Chikamai, Serestina Viriri, Jules‐Raymond Tapamo · International Journal of Computers Communications & Control · 2013
Breast cancer is among the leading cause of death among females. Studies show that early detection allows for a better prognosis. Mammography is one of the successful ways for early detection of breast cancer. It mostly involves manual reading of mammograms, a process that is difficult and error-prone. This paper discusses a classification model for mammograms based on microcalcification characteristics, as a way of helping radiologists make quick and accurate diagnostic decisions by availing to them similar past cases. The images are pre-processed by Gaussian smoothing and median filtering with 5 x 5 and 3 x 3 kernels respectively. Gabor and Haralick features are then extracted to form the image signatures over which similarity measurements are made. Experimental results show an average precision value between 0.5 and 0.61 using Haralick features, 0.49 and 0.57 using Gabor features, and 0.51 and 0.78 using combination of Gabor and Haralick features.