Fast assist for brain MRI tumor detection.

Upload a scan, run inference, and get an outline plus confidence with a PyTorch CNN trained on 4,600 MRI slices (F1 0.993, accuracy 0.993 on full set). Built to lighten your second-read workload—radiologist oversight required.

CNN · PyTorchHeatmap overlaysPDF/CSV-ready output

Model snapshot

F1 0.993

CNN in PyTorch (4 conv blocks + 2 FC), dropout 0.25
Trained on 4,600 MRI slices (tumor vs healthy)
Validation on full set: accuracy 0.993, F1 0.993 (CPU run)
Input: 3x256x256; augmentations: flips, rotation, normalization
Optimiser: Adam 3e-4 · CrossEntropyLoss

Swap in your own weights or ONNX export. For heavy inference, point the `/api/analyze` route to a GPU endpoint.

Upload & analyze

Local-only upload

Choose MRI file

DICOM or image formats · kept in browser, sent only to this app for analysis.

Input

3 x 256 x 256

Turnaround

~1–2 seconds on GPU

Output

Label, confidence, overlay

Result

Demo-friendly

Run an analysis or load a demo MRI to see predicted label, confidence, and overlay.

Successful cases

Examples with overlays

Swap in your own anonymized examples to showcase wins.

Glioma · left frontal

Detected

Trained data sample

Sample from training set; clearly hyperintense rounded mass.

Confidence 92%

Meningioma-like

Detected

Trained data sample

Extra-axial appearing mass; part of training distribution.

Confidence 95%

No tumor detected

Clear

Trained data sample

Negative example from training set; no focal lesion present.

Confidence 98%

Responsible use

Assistive tool only. Radiologist oversight required. Do not serve PHI without proper safeguards. For local-only workflows, keep inference on-device or inside your hospital network.