Final Degree Project for Computer Engineering, the project is oriented towards the diagnosis of multiple chest pathologies with deep learning, studying the effectiveness of transfer learning techniques. The datasets for the model development are acquired both by Kaggle and CheXpert, a dataset provided by Stanford ML Group, with over 200.000 samples with both frontal and lateral X-rays and 14 different observations.
First, the data is cleaned and prepared for the training process, where several architectures and techniques are tested, like transfer learning from deep convolutional neural networks , such as DenseNet, image augmentation techniques and parallel CNNs. Then, an explanation of the model is performed to detect how it decides whether a certain pathology exists or not and what are the areas of interest for each pathology, comparing its thinking process with state of the art methods used by doctors to check if the predictions have sense.
Finally, the obtained model is deployed on a web server, accesible in this repo, which can be used to upload an X-Ray to get a fast real-time analysis. An additional case motivated by the current situation has also been studied, using transfer learning to detect COVID-19 and differentiate it from usual pneumonia, reaching an accuracy higher than 90% and an AUC of 0.98.
Publication (September, 2021): ''Deep Transfer Learning for Interpretable Chest X-Ray Diagnosis'' was accepted by HAIS 2021.
In order to train the models its neccesary to acquire the datasets being used and to install all the requirements.
The datasets used for the project are the following:
- Clone the repository
git clone https://github.com/carloslago/IntelligentXray.git- Install Python packages
pip install requirements.txt- CheXpert: code for training models for the CheXpert challenge.
- Covid: COVID-19 network.
- Pneuomonia: network for the pneuomina kaggle challenge.
- XretAbnormalities: network with transfer learning from the pneumonia case.
- utils: scripts to reorder datasets csv and get network statistics.
Feel free to cite the following publication:
@article{lago_lopez-gazpio_onieva_2021, title={Deep Transfer Learning for interpretable chest X-ray diagnosis}, DOI={10.1007/978-3-030-86271-8_44}, journal={Lecture Notes in Computer Science}, author={Lago, C. and Lopez-Gazpio, I. and Onieva, E.}, year={2021}, pages={524–537}}