Segmentation-guided NSCLC survival inference
Upload a chest CT in NIfTI format. The system runs nnU-Net to generate a periskeletal mask, then a 3D ResNet-34 classifier predicts two-year survival. Grad-CAM overlays are provided for axial, coronal, and sagittal center slices.
This interface is designed for clinical research workflows and does not provide medical advice.
Model Summary
SegmentationnnU-Net v2 (3D fullres)
Classifier3D ResNet-34
InputCT + mask
Output2-year survival
Run Inference
Pipeline Steps
1) nnU-Net periskeletal segmentation -> 2) crop & normalize CT volume -> 3) ResNet-34 survival prediction -> 4) Grad-CAM heatmaps for review.
For consistent results, use scans with standard Lung1 preprocessing.
Method Highlights
Periskeletal context
Explicit anatomical priors improve interpretability and robustness.
Segmentation-guided
Mask + CT fusion constrains attention to clinically meaningful regions.
Transparent outputs
Grad-CAM overlays support qualitative review by clinicians.