Robot-Assisted Detection and Stiffness Prediction of Breast Tumor

Abstract

A methodology of robot-assisted breast tumor detection and stiffness prediction has been established. In the paper, a statistical shape model for tumor morphology was built and utilized for tumor phantom fabrication in palpation experiments. The force signals acquired from robot-held probe during the automated palpation were pre-processed to reduce the surrounding tissue effect before the subsequent detection of force peak, slope feature, and shape correlation. The tumor detection by the proposed algorithm could produce robust results for tumor existence. Subsequently, the force curve features were also extracted for training the Support Vector Regression (SVR) model for tumor stiffness prediction. The result showed that the prediction error was 8% and 15% respectively for the trained and interpolated stiffness for the testing set. The accuracy of stiffness prediction is acceptable for providing the second opinion as a companion to image modality for breast cancer diagnosis.