Generalization of chest x-ray deep learning models
Principal Investigator
Name
Santiago Ibanez Caturla
Degrees
MD
Institution
Hospital Universitario Virgen de la Arrixaca
Position Title
Radiologist
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-625
Initial CDAS Request Approval
May 18, 2020
Title
Generalization of chest x-ray deep learning models
Summary
There are many public chest x-ray datasets that allow researchers to develop their own deep learning models for both classification and segmentation. Each of them has its own features, such as different disease spectrum, patient distribution and so on.
Most of the models are usually built using just one dataset, and although performance it may not perform poorly, these models frequently lack one necessary feature: generalization, which allows models created from one dataset to be able to make predictions to other datasets.
The project wants to evaluate the effect of generalization by using several datasets to train a combined model, and compare it to single-dataset trained models.
Most of the models are usually built using just one dataset, and although performance it may not perform poorly, these models frequently lack one necessary feature: generalization, which allows models created from one dataset to be able to make predictions to other datasets.
The project wants to evaluate the effect of generalization by using several datasets to train a combined model, and compare it to single-dataset trained models.
Aims
The aims of this project are:
- Evaluate the dataset and its characteristic features
- Use the dataset to create a combined model
- Evaluate them with different metrics.
Collaborators
Oscar Martinez Mozos, PhD (Ôrebro University, Sweden)
Juan de Dios Berná, PhD (University of Murcia)
Guillermo Carbonell, PhD (University of Murcia)
Daniel Rodriguez, PhD (Hospital Universitario Virgen de la Arrixaca)