S-J Kuo1, Y-H Hsiao, Y-L Huang, D-R Chen. 1. Comprehensive Breast Cancer Center, Surgical Research Laboratory, Changhua Christian Hospital, Changhua, Taiwan.
Abstract
OBJECTIVES: To evaluate the use of three-dimensional (3D) power Doppler ultrasound in the differential diagnosis of solid breast tumors using a neural network model as a classifier. METHODS: Data from 102 benign and 93 malignant breast tumor images that had pathological confirmation were collected consecutively from January 2003 to February 2004. We used 3D power Doppler ultrasound to calculate three indices (vascularization index (VI), flow index (FI) and vascularization flow index (VFI)) for the tumor itself and for the tumor plus a 3-mm shell surrounding it. These data were applied to a multilayer perception (MLP) neural network model and we evaluated the model as a classifier to assess the capability of 3D power Doppler sonography to differentiate between benign and malignant solid breast tumors. RESULTS: The accuracy of the MLP model for classifying malignancy was 84.6%, the sensitivity was 90.3%, the specificity was 79.4%, the positive predictive value was 80.0% and the negative predictive value was 90.0%. When the neural network was used to combine the three 3D power Doppler indices, the area under the receiver-operating characteristics curve was 0.89. CONCLUSIONS: 3D power Doppler ultrasound may serve as a useful tool in distinguishing between benign and malignant breast tumors, and its capability may be increased by using a MLP neural network model as a classifier.
OBJECTIVES: To evaluate the use of three-dimensional (3D) power Doppler ultrasound in the differential diagnosis of solid breast tumors using a neural network model as a classifier. METHODS: Data from 102 benign and 93 malignant breast tumor images that had pathological confirmation were collected consecutively from January 2003 to February 2004. We used 3D power Doppler ultrasound to calculate three indices (vascularization index (VI), flow index (FI) and vascularization flow index (VFI)) for the tumor itself and for the tumor plus a 3-mm shell surrounding it. These data were applied to a multilayer perception (MLP) neural network model and we evaluated the model as a classifier to assess the capability of 3D power Doppler sonography to differentiate between benign and malignant solid breast tumors. RESULTS: The accuracy of the MLP model for classifying malignancy was 84.6%, the sensitivity was 90.3%, the specificity was 79.4%, the positive predictive value was 80.0% and the negative predictive value was 90.0%. When the neural network was used to combine the three 3D power Doppler indices, the area under the receiver-operating characteristics curve was 0.89. CONCLUSIONS: 3D power Doppler ultrasound may serve as a useful tool in distinguishing between benign and malignant breast tumors, and its capability may be increased by using a MLP neural network model as a classifier.