Breast cancer clusterization using physical parameters and support vector machine methods

Background: Early detection of breast cancer can reduce mortality due to breast cancer. Mammography is the current standard for screening, but it is less sensitive for tumors <1 mm. Computed Tomography (CT) Scan is the most accurate method to see the spreading of breast cancer. This paper aims to provide a general description of the use of physical parameters to diagnose breast cancer quickly.

Methods: We used images from Dr. Soetomo General Hospital database. The database has 284 cases with 1024x1024 pixels resolution, which consists of 102 normal and 182 abnormal mammograms. We conducted ANOVA test to choose which parameters were the most significant which were able to distinguish between normal and abnormal, which we would later use as input variables for the SVM method.

Results: The entropy parameter provided a sensitivity of 81.81%, and specificity of 88.89%. The contrast parameter obtained sensitivity of 81.81% and specificity of 88.89%. The entropy of Hdiff parameter obtained 64.29% sensitivity and 83.33% specificity.

Conclusion: The use of physic parameters for early detection of breast cancer provides high sensitivity and specificity. These physic parameters can be used as parameters to classify breast cancer.