Fig. 4: Dependence of V-EUS on tumor depth in diagnosing breast cancer.
From: Virtual elastography ultrasound via generative adversarial network for breast cancer diagnosis
a Comparison of ROCs in determining tumor malignancy on test set when dividing training and test set with 15 mm as the threshold. b The diagnostic performance of real EUS and V-EUS varies with the depth of tumor. For the real EUS, the centers of the error bar for each interval are 0.812, 0.790, 0.775, 0.709, 0.702, and 0.647, respectively. For the V-EUS, the centers of the error bar for each interval are 0.766, 0.777, 0.770, 0.781, 0.791 and 0.794 respectively. n indicates the number of cases in the interval. Error bar indicates 95% confidence intervals. (*p < 0.05; **p < 0.01, the p-values for the last three intervals are 0.0013, 0.0017, 0.0004 respectively). c Examples of typical case results. ROIs were cropped from the US images and displayed on the right together with V-EUS. We observe that for the deep-located tumor, V-EUS not only perform better than real EUS, but also avoid artifacts caused by US signal attenuation. Pink arrows highlight the US imaging at the signal attenuation. Source data are provided as a Source Data file.
