Dl based models for classifying kidney biopsies of patients with ANCA-associated glomerulonephritis according to the Berden histopathological classification system.
In the codes folder you will find codes for training and testing the classification models X1, X2 and X3 (find specifications below) and the segmentation UNet model. Also the code which was used to extract the GradCam data, is provided.
In the Images folder you will find sample patches of training data, with and without back ground. Also there are sample images of GradCAM algorithm output as well as the performance of the segmentation algorithm
| Dataset w bg | Dataset w/o bg | Pre-trained on | Architecture | Img dim |
|---|---|---|---|---|
| X1_bg | X1_nobg | Imagenet | Inceptionv3 | 150x150 |
| X2_bg | X2_nobg | Imagenet | EfficientNetB1 | 224x224 |
| X3_bg | X3_nobg | Imagenet+DPD | EfficientNetB1 | 224x224 |
DPD: Digital Pathology Dataset (refer to the publication)
Will be updated soon.
The trained models can be found in the google drive link below. https://drive.google.com/file/d/1DZ9fQteJYvHt9gjoWxQOrn1GolDCJGdw/view?usp=sharing