Code for MLHC 2021 Multi-Label Generalized Zero Shot Learning for the Classification of Disease in Chest Radiographs

Background
Overview
Environment setup
Dataset
Model evaluation
Model training
Citation
Results

Background

Despite the success of deep neural networks in chest X-ray (CXR) diagnosis, supervised learning only allows the prediction of disease classes that were seen during training. At inference, these networks cannot predict an unseen disease class. Incorporating a new class requires the collection of labeled data, which is not a trivial task, especially for less frequently-occurring diseases. As a result, it becomes inconceivable to build a model that can diagnose all possible disease classes. This repo contains the pytorch implementation for our proposed network; multi-label generalized zero shot learning (CXR-ML-GZSL) that can simultaneously predict multiple seen and unseen diseases in CXR images. Given an input image, CXR-ML-GZSL learns the visual representations guided by the input's corresponding semantics extracted from a rich medical text corpus.

Overview of the CXR-ML-GZSL network

The components of the network consist of (i) a trainable visual encoder, (ii) a fixed semantic encoder, & (iii) a projection module to map the encoded features to a joint latent space. Our approach is end-to-end trainable and does not require offline training of the visual feature encoder.

Environment setup

git clone https://github.com/nyuad-cai/CXR-ML-GZSL.git
cd CXR-ML-GZSL
conda env create -f environment.yml
conda activate zsl

Dataset

We evaluated the proposed method on the NIH Chest X-ray dataset with a random split of 10 seen and 4 unseen classes. To train and evaluate the network, download the NIH chest x-ray dataset.

Model evaluation

To perform evaluation only, you can simply download the pretrained network weights.
Update the paths of data-root directory and pretrained weights to run the following script.

sh ./scripts/test_densenet121.sh

Model training

To train the network, you run the following script by setting the data-root directory path of the downloaded dataset.

sh ./scripts/train_densenet121.sh

Citation

If you use this code for your research, please consider citing:

@misc{hayat2021multilabel,
      title={Multi-Label Generalized Zero Shot Learning for the Classification of Disease in Chest Radiographs}, 
      author={Nasir Hayat and Hazem Lashen and Farah E. Shamout},
      year={2021},
      eprint={2107.06563},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Results

We compare the performance of our proposed approach with two baseline methods and report overall precision, recall, f1-score @k {2,3} and AUROC for seen & unseen classes.

Class wise comparison with baseline methods. The last 4 italicized classes are unseen during training.

Below we show some of the visual results for top 3 predictions. Green, orange, and red represent true positives, false negatives, and false positives, respectively. Note that our method predicts the unseen classes as top-3 even when the number of ground-truth labels is greater than 3.

nyuad-cai / CXR-ML-GZSL