brain-tumor-attention-unet – AI Model by tannnnnnvvviiii | AlphaNeural AI

You can deploy this model and start earning money today!

tannnnnnvvviiii/
brain-tumor-attention-unet

like

0

medical-imaging

computer-vision

Views

No views yet

🧠 Brain Tumor Segmentation — Attention U-Net

HuggingFace Space

Model Description

Attention U-Net trained on BraTS2020 for brain tumor segmentation from multi-modal MRI (FLAIR, T1, T1ce, T2).

Results

Binary Segmentation

Metric	Score
Dice Score	0.81
AUC-ROC	0.99
Sensitivity	0.67
Specificity	0.99

Multi-class Segmentation

Region	Dice
Whole Tumor	0.82
Tumor Core	0.59
Enhancing Tumor	0.53

Architecture

Attention U-Net (31.4M parameters)
Input: 4-channel MRI (128x128)
Attention gates focus on tumor regions
Grad-CAM explainability

Files

model.pth — Binary segmentation model
multiclass_unet.pth — Multi-class model (3 regions)
confidence_heatmap.png — Prediction uncertainty visualization
3d_reconstruction.png — 3D tumor volume projection
roc_curve.png — ROC curve (AUC=0.99)