Mammographic Masses Classification: Comparison between backpropagation neural network (BNN), K nearest neighbors (KNN), and human readers

摘要

PURPOSE: We compare mammographic mass classification performance between a backpropagation neural network (BNN), K nearest neighbors, expert radiologists, and residents. Our goal is to reduce false negatives during screening of mammograms. METHODS: 160 cases were used. Each case contained at least one mass and had an accompanying biopsy result. Masses were extracted using region growing with seed locations identified by an expert radiologist. 10 texture and shape based features were used as inputs to a BNN and KNN. 140 cases were used for training the BNN and the KNN. The remaining 20 cases were used for testing. The testing set was diagnosed by three expert radiologists, three residents, the BNN, and the KNN. We evaluated the human readers and the BNN by computing the area under the ROC curve (AUC). The KNN was evaluated by computing the sensitivity, specificity, and number of false negatives (FN). RESULTS: The AUC was 0.923 for the BNN, 0.846 for the expert radiologists, and 0.648 for the residents. The KNN had a specificity 85.7% with sensitivity 84.6% and with FN=2. These results illustrate the promise of using the BNN as a physician's assistant for breast mass classification.