Computer vision models for EVLP lung X-ray images.
Welcome to the EVLP X-ray Image Classification repository. This space contains modules for data loading, model training, on-the-fly validation, and inference, specifically designed for running convolutional neural networks on ex vivo lung radiographs.
To train a model, navigate to the 'script' folder and run either the finetune script. Note that for our group's upcoming radiographic analysis paper, pretrained models were trained separately using PyTorch Image Library (timm).
Explore the 'dataset' folder, which contains various files for processing and loading different datasets. For our paper, we used the 'evlp_xray_outcome' file.
For training a trend model that takes images from different time points as input, set 'trend=True' in the finetune script. This will invoke the 'trend_model' file under the 'models' folder.
To obtain predicted probabilities and latent image features, execute the 'outcome_getprobs_getfinalfeatures.py' file under the 'inference' folder.
For visualizing class activation saliencies, run the 'saliency.py' file located in the 'inference' folder.