Curated 4-class aggregate / license review
bv-brainmr-v1 · BlackVoxel
Brain MRI tumor classification.
MRI-slice classification across no tumor, glioma, meningioma and pituitary tumor.
What it does
The model classifies one MRI slice into four categories: no tumor, glioma, meningioma or pituitary tumor. The curated dataset does not represent routine clinical complexity.
- Model
- bv-brainmr-v1
- Ownership
- BlackVoxel
- Base
- ResNet-18 / ImageNet transfer
- Data
- Curated 4-class aggregate / license review
Demo
Image, attention map and draft report.
The screen uses previously processed results. Inference is not run on this page.
Meningioma · 100%
MRW/L · AUTO
STORED MODEL OUTPUTGRAD-CAM
Training and evaluation
Data and setup.
ResNet-18 / ImageNet transfer
Task-specific labels
curated benchmark
ResNet-18 fine-tuned locally on an aggregated four-class MRI-slice dataset. The checkpoint was selected on validation and evaluated on the test set.
On the 1,055-image test set, macro-AUROC was 0.9991 and accuracy was 0.9896.
Test results
Measured performance.
macro-AUROC
curated benchmarkaccuracy
test n=1,055status
license reviewThe result comes from a curated benchmark where published baselines also reach 98% or 99%. No clinical evaluation was performed.
Limitations
Scope of this result.
- 01
The aggregate dataset license is still under review. No commercial use.
- 02
It classifies one slice, not the complete volumetric study.
- 03
Four curated classes do not represent clinical diversity.
Next model
US / BREAST