Metastasis classification (CAMELYON16)
This model classifies an H&E-stained digital pathology image of axillary lymph nodes as not containing metastasis or containing metastasis. It was trained by Jakub Kaczmarzyk.
Inputs: Bag of patches with 128um edge length, embedded with CTransPath.
Output classes: no-metastasis, metastasis
Data
CAMELYON16 was used to train the model. The whole slide images were tiled into 128x128um patches, and each patch was encoded using CTransPath (this produces 768-dimensional embeddings).
Train and validation splits were stratified by metastasis status. The test set is pre-defined in the CAMELYON16 dataset.
Samples sizes:
- Train: 243 slides
- Validation: 27 slides
- Test: 129 slides
Reusing this model
To use this model on the command line, see
WSInfer-MIL.
Alternatively, you may use PyTorch on ONNX to run the model. First, embed 128um x 128um patches using CTransPath. Then pass the bag of embeddings to the model.
1import onnxruntime as ort
2import numpy as np
3embedding = np.ones((1_000, 768), dtype="float32")
4ort_sess = ort.InferenceSession("model.onnx")
5logits, attention = ort_sess.run(["logits", "attention"], {'input': embedding})
Model performance
The model achieves an AUC of 0.91 in the test set of CAMELYON16. Below, please find a confusion matrix, where predicted classes are columns, and true values are rows.
| No-Met | Met |
|---|
| No-Met | 77 | 3 |
| Met | 12 | 37 |
Intended uses
This model is ONLY intended for research purposes.
This model may not be used for clinical purposes. This model is distributed without warranties, either express or implied.
