Breast Cancer Diagnosis Using Shape Analysis and Decision Tree Technique

ABSTRACT

Breast cancer is one of the leading cause of death among women in Nigeria and the world at large. The increase rate of this type of cancer called for more facilities for early detection. The primitive methods (self-examination, Hormone Replacement Therapy (HRT), X-ray mammogram) of its detection and treatment in the past seems not be enough thereby not improve the survival rate. There is a need to reduce the mortality rate by proper diagnosis of this disease using computerized imaging technique to reduce the false positive and false negative results generated from the traditional x-ray film use. This research work focuses on using various geometric shapes and margin features to differentiate benign and malignant masses. 16 geometrical shapes and margin features are introduced out of which 9 features were selected to characterize the masses and margin features. We conducted experiment on 114 mammogram images gotten from adult participants visiting a screening center in Victoria Island, Lagos for a routine checkup. The experimental result shows that Classification And Regression Tree analysis (CART) can be used to distinguish benign masses from malignant masses effectively and generates simple rules, which can be easily implemented on any system using IF..THEN..ELSE statements. Results generated with CART Algorithm are found to be encouraging with 97.1% accuracy. When compared with other methods, such as Support Vector Machine with Gaussian Kernel which gives 81% accuracy and Genetic Algorithm based feature selection with Neural Network gave 85% accuracy. Our results show that CART performs better than the existing algorithms used for classifying masses as benign or malignant.

INDEX TERMS Breast cancer, Decision Tree, Machine learning, feature selection, masses