You can deploy this model and start earning money today!

kssrikar4/
KidneySeg-UNet

Name: KidneySeg-UNet
Availability: InStock

mpl-2.0

Views

No views yet

KiTS Kidney Tumor Segmentation

Two-stage segmentation pipeline for kidney and tumor delineation in abdominal CT scans, optimized for the KiTS Challenge. Combines coarse volumetric localization with high-resolution refinement using MONAI's SegResNet.

Architecture

Stage	Purpose	Input	Output
1 (Localization)	Coarse kidney ROI extraction	128³ downsampled volume	Binary kidney mask
2 (Refinement)	Fine-grained kidney + tumor segmentation	Native-res crop + Stage 1 mask (3 channels)	Multi-class segmentation (kidney/tumor)

Training Configuration

Loss: DiceFocalLoss (γ=2.0) – mitigates class imbalance, emphasizes boundary accuracy
Optimizer: AdamW + CosineAnnealingLR
Augmentations: RandRotated, RandFlipd, RandGaussianNoise, RandGaussianSmooth, RandScaleIntensity, RandShiftIntensity
Framework: PyTorch + MONAI

Results

Evaluated on held-out test set (Dice similarity coefficient):

Stage	Best Epoch	Dice Score
1 (Localization)	66	0.87
2 (Refinement)	110	0.66

Note: Stage 2 operates on high-resolution, tumor-sparse crops cause boundary precision is critical. The lower Dice reflects the inherent difficulty of fine-grained tumor delineation—not model failure.

Results on Test Set

Code & Usage

The full training and inference code, along with a Streamlit‑based graphical user interface for running the model on your own CT/CBCT volumes, is available in the GitHub repository.

License

Mozilla Public License Version 2.0 - Feel free to use and modify

kssrikar4/KidneySeg-UNet

KiTS Kidney Tumor Segmentation

Architecture

Training Configuration

Results

Results on Test Set

Code & Usage

License

kssrikar4/
KidneySeg-UNet