BUSI / CC BY 4.0
bv-breastus-v1 · BlackVoxel
Breast ultrasound classification.
Classification across normal, benign nodule and malignant nodule on BUSI.
What it does
The model classifies one B-mode frame as normal, benign or malignant. Medical assessment uses the complete sweep.
- Model
- bv-breastus-v1
- Ownership
- BlackVoxel
- Base
- ResNet-18 / ImageNet transfer
- Data
- BUSI / CC BY 4.0
Demo
Image, attention map and draft report.
The screen uses previously processed results. Inference is not run on this page.
Suspicious nodule · 86.2%
B-MODEW/L · AUTO
STORED MODEL OUTPUTGRAD-CAM
Training and evaluation
Data and setup.
ResNet-18 / ImageNet transfer
Task-specific labels
three classes
ResNet-18 fine-tuned on BUSI with three classes. The dataset has 780 images from one center, with 118 images in the test set.
Macro-AUROC 0.972, AUROC 0.980 for malignant versus the other classes and test accuracy 0.864.
Test results
Measured performance.
macro-AUROC
three classesmalignant AUROC
versus other classesaccuracy
test n=118BUSI contains images from one center. The evaluation does not measure variation across equipment, operators or populations.
Limitations
Scope of this result.
- 01
Single image, not the complete sweep.
- 02
Single-center dataset without external validation.
- 03
Grad-CAM is coarse attention, not a lesion contour or measurement.
Next model
CT / HEAD