Pretrained models for classification, segmentation and detection of different radiological conditions from chest X-ray images
This is a package with pretrained models to interpret different radiological conditions from chest X-ray images.
To use the code, first clone the repo using:
git clone https://github.com/hasibzunair/cxr-predictor.git
Navigate to the project folder and create a new working Python 3.6 environment with conda or pip. Then install the following dependencies.
pip install -r requirements.txt
You are all setup!
TODO.
Here's the list of pretrained models that are available. Training details/codes will be added later.
| Model | Dataset | Image Size | Checkpoint |
|---|---|---|---|
| DenseNet121 | CheXpert | 224x224 | CheXpert_DenseNet121_res224.h5 |
| EfficientNet B1 (featurewise std) | NIH | 224x224 | NIH_EfficientNetB1_res224.h5 |
| EfficientNet B1 (0-1 normalization) | NIH | 224x224 | NIH_EfficientNetB1_res224_rescale01.h5 |
| Mask RCNN | RSNA Pneumonia Detection | 224x224 | RNSA_Pneumonia_MaskRCNN_7epochs.h5 |
| U-Net | SIIM-ACR Pneumothorax Segmentation | 224x224 | SIIM-ACR_UEfficientNetB4_res256.h5 |
| U-Net with SWA | SIIM-ACR Pneumothorax Segmentation | 224x224 | SIIM-ACR_UEfficientNetB4_SWA_res256.h5 |
Demo notebooks are available in notebooks/
Datasets used to build the models