AI model for wound tissue classification - Red/Yellow/Black tissue detection
During my years as a registered nurse, wound assessment was part of my daily routine. We documented tissue types using TIME reporting — tracking the percentage of red, yellow, and black tissue in a wound to monitor healing progress.
The problem I kept running into: it was highly subjective. A morning nurse would document 80% red tissue. The evening nurse would look at the same wound and write 50%. Same patient, same wound, different eyes.
That inconsistency stuck with me. When I started studying AI, this felt like an obvious problem to tackle — can a model look at a wound photo and give a consistent, objective tissue breakdown that doesn't change depending on who's on shift?
This project is my attempt at that.
Nika de Vries — Applied Data Science & AI, De Haagse Hogeschool
Former registered nurse, now studying AI.
Upload a wound photo. The model finds the wound, analyses the tissue colours inside it, and returns a percentage breakdown.
| Tissue | Colour | Clinical meaning |
|---|---|---|
| Granulation | 🔴 Red | Healthy, healing tissue |
| Fibrin/Slough | 🟡 Yellow | Stalling or infected |
| Necrosis | ⚫ Black | Dead tissue |
This maps directly onto the T (Tissue) component of TIME wound assessment — the same framework I used at the bedside.
Trained on 986 wound images. Validation Dice score: 0.8861 after 20 epochs. Architecture: U-Net + ResNet34 (PyTorch).
Open in Google Colab and run the cells top to bottom. Upload your own image in the last cell.
pip install segmentation-models-pytorch albumentations torchDataset: Wang et al. (2020). Fully Automatic Wound Segmentation
with Deep Convolutional Neural Networks. Scientific Reports.
https://doi.org/10.1038/s41598-020-78799-w
Model architecture via segmentation-models-pytorch
https://github.com/qubvel/segmentation_models.pytorch
The HSV-based tissue colour analysis is my own addition.
Research project built as part of my AI studies. Not validated for clinical use. Do not use for medical decisions.