Open Source

AI-Powered Wound
Segmentation

Binary segmentation model that classifies wound vs. non-wound pixels in medical images. Built for reproducibility, security, and clinical decision support.

GitHub (private — available upon request)CLI Reference

Input

Segmentation

ClassificationWOUND
0.87

94.2%

Dice Score

0.55

Threshold

256px

Resolution

Secure

Safe Loading

ONNX

Export

Features

U-Net training, threshold calibration, 3-state classification, and ONNX export.

Explore features →

Architecture

U-Net encoder-decoder with skip connections. Modular, secure design.

View architecture →

CLI Reference

Complete command-line interface for training, inference, and export.

See CLI docs →

3-State Clinical Classification

Designed for high-stakes clinical decisions where uncertainty should be flagged for human review.

WOUND

High confidence wound detection above operating threshold

UNCERTAIN

Borderline cases flagged for human clinical review

NO WOUND

High confidence no wound below rejection threshold

Built With

PythonPyTorchU-NetONNXPandasMatplotlibNumPyPydanticClick CLI

Ready to train your own models?

WoundSeg is open source and ready to use. Clone the repository and start training.

GitHub (private — available upon request)Read the Docs