- dp_getdata.ipynb: Prepare training data for the following steps
- dp_model.ipynb: construct and train multi-scale model, provide metrics in patches level
- dp_predict.ipynb: detect cancerous cell through sliding window, draw heatmap and privide metrics in pixel level
Tensorflow 2.0, Scikit-learn, openslide
We use this pipeline to process our data and detect cancer cells. We add morphological transformation in our pipeline to improve the performance on pixel level
We use this model to solve the challenge of different scale.
In pixel level, we use IOU, accuracy and recall to evulate our model
Image | IOU | Accuracy | Recall |
---|---|---|---|
110 | 0.70 | 0.89 | 0.93 |
101 | 0.55 | 0.98 | 0.70 |
091 | 0.42 | 0.82 | 0.98 |
In patches, we use ROC and AUC to evulate our result