Step 01
Upload optional model
Drop a trained .keras or .h5 file into the notebook to swap demo probabilities for real predictions.
Dashboard outputs
Every run leaves a defensible paper trail for teammates, professors, or clinicians reviewing the pipeline.
Prediction cards
Cancer vs Not Cancer labels, confidence bars, and filename overlays rendered in a cinematic gallery.
Dark gallery
Lesion tiles are aligned, labeled, and exported as pred_gallery.png for quick sharing.
Risk strip
Top 3 lesions sorted by malignant probability with warning borders and percent callouts.
Summary panel
Pie chart + histogram showing benign vs malignant counts and the probability distribution.
Structured CSV
predictions.csv lists filename, class, malignant %, and benign % for downstream QA.
ZIP evidence
Cards, charts, and CSV are bundled into skin_cancer_outputs.zip with a manual download button.
Latest report snapshot
Prediction table
| Image | Prediction | Malignant % | Benign % |
|---|---|---|---|
| lesion_A.png | Cancer | 82.6% | 17.4% |
| lesion_B.png | Not Cancer | 31.8% | 68.2% |
| lesion_C.png | Cancer | 64.3% | 35.7% |
| lesion_D.png | Not Cancer | 18.9% | 81.1% |
Report callouts
- Average malignant confidence hovered at 60.1%.
- ZIP delivery keeps reviewers from hunting through multiple tabs.
- Grad-CAM overlay toggles on automatically if the uploaded model architecture exposes convolutional maps.
- All visuals render with the same neon-dark styling as the main site for branding consistency.
Workflow
Step 02
Load 3–8 lesion images
Images are resized to 224×224, normalized, and queued for inference or deterministic mock scoring.
Step 03
Generate deliverables
Gallery, risk strip, summary, CSV, individual cards, and a ZIP bundle are rendered with timestamped headers.
Online compiler friendly
Run the Colab-ready script
Launch Google Colab, choose Upload, and paste the script below into a new notebook cell. Everything else—UI, downloads, and report visuals—is handled for you.
Loading script…