Skip to main content
. Author manuscript; available in PMC: 2023 Dec 1.
Published in final edited form as: Med Image Anal. 2023 Sep 14;90:102960. doi: 10.1016/j.media.2023.102960

Table 5.

Classification performance of the proposed method and comparison methods on the fully-labeled BUSI and F-MBUD datasets.

Dataset Method Accuracy Precision Recall F1-Score AUC p-value
BUSI ViT 0.878 ± 0.030 0.884 ± 0.014 0.878 ± 0.030 0.877 ± 0.032 0.953 ± 0.019 ***
EB0 0.919 ± 0.017 0.920 ± 0.017 0.919 ± 0.017 0.919 ± 0.017 0.961 ± 0.007 ***
R18 0.886 ± 0.017 0.899 ± 0.030 0.886 ± 0.017 0.885 ± 0.017 0.964 ± 0.012 ***
R50 0.939 ± 0.001 0.940 ± 0.003 0.939 ± 0.001 0.939 ± 0.001 0.967 ± 0.004 ***
AGN 0.898 ± 0.047 0.901 ± 0.042 0.898 ± 0.047 0.898 ± 0.047 0.933 ± 0.020 –
R18+RMTL 0.935 ± 0.017 0.939 ± 0.019 0.935 ± 0.017 0.935 ± 0.017 0.965 ± 0.011 ***
R50+RMTL 0.955 ± 0.017 0.959 ± 0.014 0.955 ± 0.017 0.955 ± 0.017 0.982 ± 0.010 *
R18+LA-Net (Ours) 0.951 ± 0.001 0.955 ± 0.006 0.951 ± 0.001 0.951 ± 0.001 0.980 ± 0.005 ***
R50+LA-Net (Ours) 0.964 ± 0.001 0.964 ± 0.003 0.964 ± 0.001 0.964 ± 0.001 0.989 ± 0.004 –
F-MBUD ViT 0.905 ± 0.026 0.913 ± 0.075 0.889 ± 0.039 0.894 ± 0.014 0.970 ± 0.007 *
EB0 0.881 ± 0.026 0.879 ± 0.031 0.859 ± 0.031 0.867 ± 0.029 0.915 ± 0.018 ***
R18 0.893 ± 0.001 0.880 ± 0.001 0.902 ± 0.017 0.887 ± 0.003 0.923 ± 0.005 ***
R50 0.911 ± 0.001 0.901 ± 0.004 0.904 ± 0.018 0.902 ± 0.007 0.954 ± 0.009 **
AGN 0.890 ± 0.030 0.891 ± 0.030 0.890 ± 0.030 0.890 ± 0.030 0.938 ± 0.016 ***
R18+RMTL 0.833 ± 0.026 0.818 ± 0.027 0.830 ± 0.030 0.822 ± 0.026 0.910 ± 0.011 ***
R50+RMTL 0.887 ± 0.026 0.876 ± 0.027 0.905 ± 0.032 0.882 ± 0.026 0.907 ± 0.014 ***
R18+LA-Net (Ours) 0.939 ± 0.026 0.929 ± 0.015 0.915 ± 0.016 0.921 ± 0.003 0.979 ± 0.012 –
R50+LA-Net (Ours) 0.946 ± 0.001 0.947 ± 0.002 0.935 ± 0.001 0.941 ± 0.001 0.969 ± 0.001 –

Notes: Two different networks, ResNet18 and ResNet50, were used as the FEXs in our method. The input image size was 256 × 256 pixels. The results are represented as a mean value with a ±95% confidence interval. The p-value shows the t-test result of comparing the AUC of a comparison method with our method’s. The null hypothesis of the AUC of the comparison method is larger than or equal to our method’s. The meaning of the p-values is the same as described in Table 3.