Similarity measurement on digital mammogram classification

Abstract

Breast cancer is one of the dominant causes of death in the world. Mammography is the standard for early detection of breast cancer.
In examining mammograms, the overall parenchyma pattern of the left and right breast was placed side by side for symmetry assessed of left and right breast tissue by radiologist. Thus, in building computer-aided diagnosis (CAD) system for screening mammography, it is necessary to adapt the working procedure of the radiologist. In this study, 30 training images and 30 testing images from Kotabaru Oncology Clinic in Yogyakarta were used. The first step was to enhance the image quality with median filter and contrast limited adaptive histogram equalization (CLAHE). Then, feature extraction was processed by histogram-based and by gray level co-occurrence matrix (GLCM) based. Furthermore, the similarity measurement process was used to measure the difference value between selected features, i.e. angular second moment (ASM), inverse difference moment (IDM), contrast, entropy based GLCM, and energy, on the left and right mammograms. This process was intended to assess the symmetry of left and right mammograms as radiologists do in mammography screening. The obtained results of the classification between normal and abnormal images with backpropagation algorithm were accuracy of 0.933, sensitivity of 0.833, and specificity of 1.000.