Automatic Breast Tumor Segmentation using Hierarchical K-means on Mammogram

Breast cancer is the leading common types of cancer among women. According to Basic Health Research 2013, the prevalence of cancer in Indonesia is 1.4 out of 1,000 residents or about 347,000 people. It is estimated that in 2025, the number of people who die from cancer rises to 11.5 million if no prevention and control measures are taken effectively. Prevention of breast tumors can be prevented by knowing in advance before it is too late to become cancer. Early knowledge can be done by doing breast screening using mammography. The need for rapid and objective mammography readings is needed in this period to help doctors diagnose. The purpose of this study is to present a new application to help detect breast tumors automatically. Detecting breast tumors quickly is not an easy thing. It takes a qualified radiologist who expertise can make objective decisions based on mammogram images. We use valley tracing to get the optimal number of clusters, using hierarchical k-means clustering to obtain clustering results and contiguous regions for labeling components. Our experiment showed So system testing using data testing found error of 61.1% of 36 data and truth only 38.8%.