A breast tumor classification method based on ultrasound BI-RADS data mining

In this paper, to reduce the response time of computer-aided diagnostic (CAD) systems, we proposed a feature selection algorithm that utilizes BI-RADS which is the standard clinical considerations for radiologists to illustrate the visual characteristics of breast tumors. We first apply the association rule mining technique to the medical database annotated with BI-RADS lexicons by doctors, to find out the interesting BI-RADS lexicon values. Then, we select the image processing algorithms which effectively represent the chosen BI-RADS lexicon values. Finally, the features obtained from the selected image processing algorithms are used to build our classifier using Support Vector Machine (SVM) to predict whether each tumor is benign or malignant. Our experimental result shows that our classifier is accurate with fast execution time.