Skip to main content
. 2023 Jun 5;10(3):034505. doi: 10.1117/1.JMI.10.3.034505

Table 4.

ResNet-50, DenseNet-121, and Inception V3 B-Mode patches average cross-validation patient-level results after hyperparameter tuning using data from 397 training/validation patients. The highest AUROC, sensitivity, and specificity are bolded.

Model Results reported AUROC Sensitivity Specificity
ResNet-50 Patch level 0.726 ± 0.027 0.783 ± 0.070 0.720 ± 0.044
Patient level 0.829 ± 0.036 0.816 ± 0.071 0.842 ± 0.054
Inception V3 Patch level 0.722 ± 0.017 0.697 ± 0.097 0.747 ± 0.099
Patient level 0.825 ± 0.030 0.826 ± 0.089 0.824 ± 0.105
DenseNet-121 Patch level 0.727 ± 0.021 0.710 ± 0.079 0.743 ± 0.071
Patient level 0.824 ± 0.032 0.827 ± 0.061 0.822 ± 0.084