Try the analyzer on real scans.
Drag any of the 105 ultrasound images into the analysis station, or select one. You will see the ensemble's prediction, each model's vote, the alert, and whether the prediction matched the expert label.
Sample scans
BUSI dataset · 105 imagesMalignant
| Model (weight) | Benign | Normal | Malignant |
|---|
These results were produced by running the published models from the GitHub repository (Random Forest, MLP and SVM, weighted 60/25/15) on 105 distinct images drawn from the public Breast Ultrasound Images (BUSI) dataset by Al-Dhabyani et al. Images that duplicate another scan, or whose expert outline misses a visible lesion, were swapped for other scans; every tumor outline comes from the dataset's expert annotations. The model outputs are shown exactly as produced: 93 of the 105 match the expert label, including the misses. The same dataset was used to train the models, so these samples demonstrate behavior rather than serve as an independent test.