In this experiment, we explored several deep learning base models, including AlexNet, DenseNet-121, EfficientNetV2-S, GoogLeNet, MobileNetV2, ResNet-34, and ShuffleNet, for breast cancer classification using SWE images. These base models were trained on generated SWE images from three public datasets, i.e. BUSI, Breast-Lesions-USG, and BUS-BRA. As a result, the top three models were ResNet-34, MobileNetV2, and GoogLeNet, which can be selected from "modes" list in the tool page.

Below is the usage guide for our tool:

Step 1. Click the "Upload" button to upload the SWE image to be analyzed (or click the "Example" button to load the sample SWE image).
Step 2. Click the "Submit" button to run the model and obtain the prediction results.
Step 3. Results will appear in the "Result" section after ~15s.

The three model weights are provided as .pth files, packaged together in weights.zip. Users can download this zip file directly from this site.