Fig 1.
Workflow: (a) data collection, (b) network training, and (c) network validation.
Fig 2.
Data Annotation: Radiologist-Segmented Lesion ROIs. a. The first column shows the original ultrasound images. b. The second column displays the images with integrated lesion ROI (Region of Interest) results.
Fig 3.
a. MN Block is used in the network. b. Model Architecture Diagram.
Table 1.
The model performance of the baseline.
Fig 4.
Testing results. The first row a, b, c represents the internal validation results in Baseline, Data Augmentation, and Addition of Lesion ROI, respectively. The second row d, e, f represents the external validation results in Baseline, Data Augmentation, and Addition of Lesion ROI, respectively.
Table 2.
The model performance on enhancement dataset.
Table 3.
The model performance on dataset with lesion ROI.
Fig 5.
The Grad-CAM visualizations. a : Original data; b, c, d: Grad-CAM results under original data, data augmentation, and addition of lesion ROI processing methods, respectively.