Classification of Breast Cancer Tumors using a Random Forest on Mammogram Images

Abstract

Mammography can detect lumps in early-stage breast cancer when the tumors are small and cannot be felt when touched. Mammography is an X-ray of breast tissue and the diagnosis has limitations because it has low contrast and noise. We proposed a computer vision method to classify mammogram images to reduce the visual limitations of images and doctor's subjectivity. The classification process is carried out by recognizing and processing mammographic images and analyzing them using the random forest method to obtain appropriate knowledge. Before being classified, the image is enhanced by changing the format, rotated, cropped, enlarged the contrast using contrast stretching, and removed noise using a Gaussian filter. The enhanced image was extracted using the Gray Level Co-occurrence Matrix (GLCM) method. Classification of breast cancer based on mammogram images with the random forest algorithm gave an accuracy of 70.8%, a precision of 85.7%, and a recall equal to 25%.