Chest-X Ray with Radiologist AI

A deep learning system put into web production in order to supply radiological X-ray imaging assistance to physicians.

WARNING:The application web supplied is intended to be used as reference webpage. It is currently at research stage and not yet intended as production-ready webpage. We are currently trying to improve the results of the SOTA in order to have a useful and reliable application for the chest x ray imaging diagnostic.

---

How to use it?

Project description

The web application is based on four main menus with the objective of providing medical assistance regarding to chest-X-Ray image pathology detection. The application have four different menus, each one with its corresponding functionality.

1. Detection of pneumonia : from a chest X-ray image it is possible to detect if the X-ray contains pneumonia or is a normal control.
2. Detection of multi-class pathologies: based on a radiographic image, possible pathologies are alerted.
3. Automatic medical report generation: a diagnostic report is generated in relation to possible pathologies found in the image.
4. Radiology assistant: from an X-ray image it generates the three diagnoses mentioned above.

Code execution

streamlit run app.py

---

Installation

Instructions

Run the silent installation of Miniconda/Anaconda in case you don't have this software in your environment.

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -p $HOME/miniconda3

Once you have installed Miniconda/Anaconda, create a Python 3.7 environment.

conda create --name cxr-rai python=3.8.13
conda activate cxr-rai

Clone this repository and install it inside your recently created Conda environment.

git clone https://github.com/Rules99/Chest-X-Ray-with-Radiologist-AI
cd Chest-X-Ray-with-Radiologist-AI
pip install -r requirements.txt

Dependencies

• python 3.8.13
• efficientnet 1.1.1
• gensim 3.8.3
• googletrans 4.0.0-rc1
• grad-cam 1.3.7
• h5py 3.1.0
• imgaug 0.4.0
• matplotlib 3.5.1
• nltk 3.4.5
• numpy 1.19.5
• opencv-python-headless
• pandas 1.4.2
• plotly 5.8.0
• requests 2.27.1
• scikit_image 0.19.2
• scikit_learn 1.0.2
• seaborn 0.11.2
• streamlit 1.8.1
• streamlit-option-menu 0.3.2
• tensorflow 2.5.3
• termcolor 1.1.0
• torch 1.11.0
• torchsummary 1.5.1
• torchvision 0.12.0
• torchxrayvision 0.0.32
• transformers 2.5.1
• tqdm 4.64.0
• Pillow 9.1.0
• protobuf 3.19.0

---

Authors & Contributors

The application was developed by:

• Pablo Reyes
• Fernando Pozo

---

Acknowledgements

We would like to thank the creators of the Torchxrayvision platform for sharing their pre-trained X-ray image models. Also thanks to Omar-Mohamed for reproducing automatic report generation model.

---

Repositories

• Torchxrayvision repo: torchxrayvision
• Automatic Generation repo : GPT2-Chest-X-Ray-Report-Generation

---

Citation

@misc{10.1093/nargab/lqab044,
    author = {Reyes, Pablo and Pozo, Fernando},
    title = "{Sistema de identificación e interpretación de patologías pulmonares a partir de imágenes rayos X mediante Aprendizaje Profundo}",
    year = {2022},
    month = {06},
    abstract = "{}",
    url = {}
}

---

References

Databases

• Chexpert : https://stanfordmlgroup.github.io/competitions/chexpert/
• NIH : https://www.nih.gov/
• MIMIC-CXR-JPG : https://physionet.org/content/mimic-cxr-jpg/2.0.0/
• China National Center for Bioinformation: https://ngdc.cncb.ac.cn/news/85

Papers

https://link.springer.com/article/10.1007/s12559-020-09787-5

https://www.researchgate.net/publication/355841287_TorchXRayVision_A_library_of_chest_X-ray_datasets_and_models

https://pubmed.ncbi.nlm.nih.gov/32864270/

https://www.sciencedirect.com/science/article/pii/S2352914821000472