Breast cancer diagnosis from fine-needle aspiration using supervised compact hyperspheres and establishment of confidence of malignancy

An automatic classification methodology is implemented to analyze breast masses from fine-needle aspiration, including feature selection, diagnostic decision, and computation of confidence of malignancy for each mass. Feature selection is performed using a genetic algorithm based on a measure of alignment. A kernel-based classifier is employed for diagnostic decision, which learns two supervised compact hyperspheres (SCHs), each encompassing the most training patterns from one class, while also the least training patterns from the other class. Instead of only providing a binary diagnostic decision of “malignant” or “benign”, we assign a confidence of malignancy to each mass for the first time, by calculating probabilities of being benign and malignant. Compared with several well-known classifiers, the SCH-based classifier provides the highest training accuracy of 100% and test accuracy of 99%.