Parameter	Value
Batch Size	8
Learning Rate	7.77451918775676e-06
Weight Decay	0.00164040349077736
Epochs	100
Patience	10
Dataset	GleghornLab/Semi-Automated_LN_Segmentation_10_11_2025

Validation Performance Metrics (Pooled Pixel-Level)

Metric	Mean	Class 0	Class 1	Class 2	Class 3
Dice	0.8697	0.7635	0.8607	0.9085	0.9461
IoU	0.7757	0.6174	0.7554	0.8323	0.8977
F1	0.8697	0.7635	0.8607	0.9085	0.9461
MCC	0.8649	0.7636	0.8588	0.9032	0.9340
ROC AUC	0.9971	0.9946	0.9983	0.9978	0.9978
PR AUC	0.9412	0.8527	0.9494	0.9720	0.9908

Test Performance Metrics (Pooled Pixel-Level)

Metric	Mean	Class 0	Class 1	Class 2	Class 3
Dice	0.7960	0.7578	0.7830	0.7344	0.9087
IoU	0.6666	0.6101	0.6433	0.5803	0.8328
F1	0.7960	0.7578	0.7830	0.7344	0.9087
MCC	0.7959	0.7609	0.7891	0.7436	0.8900
ROC AUC	0.9942	0.9949	0.9975	0.9903	0.9941
PR AUC	0.9117	0.8628	0.9133	0.8985	0.9720

Usage

python

import numpy as np
from model import MODEL_REGISTRY, SegmentationConfig

# Load model
config = SegmentationConfig.from_pretrained("aholk/LN_segmentation_sweep")
model = MODEL_REGISTRY["unet"].from_pretrained("aholk/LN_segmentation_sweep")
model.eval()

# Run inference on a full image with sliding window
image = np.random.rand(2048, 2048, 3).astype(np.float32)  # Your image here
probs = model.predict_full_image(
    image,
    dim=128,
    batch_size=16,
    device="cuda"  # or "cpu"
)
# probs shape: (num_classes, H, W) with values in [0, 1]

# Threshold to get binary masks
masks = (probs > 0.5).astype(np.uint8)

Citation

If you use this model, please cite:

bibtex

@software{windowz_segmentation,
  title={Multilabel Image Segmentation with Sliding Window U-Net},
  author={Gleghorn Lab},
  year={2025},
  url={https://github.com/GleghornLab/ComputerVision2}
}

GleghornLab/lymph_node_segmentation

LN_segmentation_sweep

Model Description

Model-Specific Parameters

Training Configuration

Validation Performance Metrics (Pooled Pixel-Level)

Test Performance Metrics (Pooled Pixel-Level)

Usage

Citation

GleghornLab/
lymph_node_segmentation