ensemble-pneumonia-detection's Introduction

Ensemble-Pneumonia-Detection

Official implementation of our paper titled "Pneumonia Detection from Chest X-ray Images using a Novel Weighted Average Ensemble Model" published in Nature- PLoS One.

Requirements

To install the dependencies, run the following using the command prompt:

pip install -r requirements.txt

Running the codes on Pneumonia data

The two datasets used in the paper can be found:

Dataset by Kermany et al.: https://www.kaggle.com/paultimothymooney/chest-xray-pneumonia
RSNA Pneumonia Detection challenge dataset: https://www.kaggle.com/c/rsna-pneumonia-detection-challenge

Required Directory Structure:

+-- data
|   +-- .
|   +-- train
|   +-- val
+-- train_csv
|   +-- .
|   +-- densenet121_train.csv
|   +-- googlenet_train.csv
|   +-- resnet18_train.csv
+-- test_csv
|   +-- .
|   +-- densenet121_test.csv
|   +-- googlenet_test.csv
|   +-- resnet18_test.csv
+-- main.py
+-- probability_extraction
+-- train_labels.csv
+-- test_labels.csv

To extract the probabilities on the dataset using the different models run probability_extraction.py and save the files according to the folder structure shown above.

Next, to run the ensemble model on the base learners run the following:

python main.py --root_train "train_csv/" --root_test "test_csv/" --train_labels "train_labels.csv" --test_labels "test_labels.csv"

Citation

If you use this repository, please consider citing our paper:

Kundu R, Das R, Geem ZW, Han GT, Sarkar R. Pneumonia detection in chest X-ray images using an ensemble of deep learning models. PLoS One. 2021 Sep 7;16(9):e0256630. doi: 10.1371/journal.pone.0256630. PMID: 34492046.

Bibtex:

@article{kundu2021pneumonia,
  title={Pneumonia detection in chest X-ray images using an ensemble of deep learning models},
  author={Kundu, Rohit and Das, Ritacheta and Geem, Zong Woo and Han, Gi-Tae and Sarkar, Ram},
  journal={PloS one},
  volume={16},
  number={9},
  pages={e0256630},
  year={2021},
  publisher={Public Library of Science San Francisco, CA USA}
}

Recommend Projects