Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- computational-pathology
|
| 5 |
+
- colorectal-cancer
|
| 6 |
+
- multiple-instance-learning
|
| 7 |
+
- whole-slide-image
|
| 8 |
+
- medical-imaging
|
| 9 |
+
language: en
|
| 10 |
+
datasets:
|
| 11 |
+
- custom
|
| 12 |
+
pipeline_tag: image-classification
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# CRC-ESD: Metastasis Risk Prediction Model
|
| 16 |
+
|
| 17 |
+
Trained model weights for **"Deep Learning-Based Analysis of H&E-Stained Histopathological Images for Predicting Metastasis Risk After Endoscopic Submucosal Dissection in Colorectal Cancer"**.
|
| 18 |
+
|
| 19 |
+
## Model Description
|
| 20 |
+
|
| 21 |
+
- **Architecture**: FeatherSlideEncoder (gated attention ABMIL) + 2-layer MLP classifier
|
| 22 |
+
- **Input**: UNI-2h patch features (1536-d → 512-d embedding) from H&E WSIs
|
| 23 |
+
- **Output**: Binary metastasis risk probability
|
| 24 |
+
- **Training**: 5-fold stratified cross-validation on 113 CRC patients (294 WSIs)
|
| 25 |
+
- **Performance**: Mean AUC = 0.759 ± 0.041
|
| 26 |
+
|
| 27 |
+
## Files
|
| 28 |
+
|
| 29 |
+
| File | Description |
|
| 30 |
+
|------|-------------|
|
| 31 |
+
| `best_model_fold0.pth` | Fold 0 best checkpoint |
|
| 32 |
+
| `best_model_fold1.pth` | Fold 1 best checkpoint |
|
| 33 |
+
| `best_model_fold2.pth` | Fold 2 best checkpoint |
|
| 34 |
+
| `best_model_fold3.pth` | Fold 3 best checkpoint |
|
| 35 |
+
| `best_model_fold4.pth` | Fold 4 best checkpoint |
|
| 36 |
+
|
| 37 |
+
## Usage
|
| 38 |
+
|
| 39 |
+
```python
|
| 40 |
+
import torch
|
| 41 |
+
from trident.slide_encoder_models import FeatherSlideEncoder
|
| 42 |
+
|
| 43 |
+
# Load model
|
| 44 |
+
model = FeatherSlideEncoder()
|
| 45 |
+
classifier = torch.nn.Sequential(
|
| 46 |
+
torch.nn.Linear(512, 256),
|
| 47 |
+
torch.nn.ReLU(),
|
| 48 |
+
torch.nn.Dropout(0.5),
|
| 49 |
+
torch.nn.Linear(256, 1)
|
| 50 |
+
)
|
| 51 |
+
|
| 52 |
+
checkpoint = torch.load("best_model_fold0.pth", map_location="cpu")
|
| 53 |
+
model.load_state_dict(checkpoint["encoder_state_dict"])
|
| 54 |
+
classifier.load_state_dict(checkpoint["classifier_state_dict"])
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
## Citation
|
| 58 |
+
|
| 59 |
+
```bibtex
|
| 60 |
+
@article{zhao2025crc_esd,
|
| 61 |
+
title={Deep Learning-Based Analysis of H&E-Stained Histopathological Images for Predicting Metastasis Risk After Endoscopic Submucosal Dissection in Colorectal Cancer},
|
| 62 |
+
author={Zhao, Zhixun and Gong, Changhao and Shi, Yihang and others},
|
| 63 |
+
journal={npj Digital Medicine},
|
| 64 |
+
year={2025}
|
| 65 |
+
}
|
| 66 |
+
```
|
| 67 |
+
|
| 68 |
+
## Code
|
| 69 |
+
|
| 70 |
+
GitHub: [https://github.com/ChanghaoGong/CRC-ESD](https://github.com/ChanghaoGong/CRC-ESD)
|