Probabilistically localize suspected foci of pneumonia on frontal chest radiographs.

Intended Use

Any PA/AP Chest Radiograph

Limitations

Does not consider prior information for the same patient as part of a prior prior likelhood.

Information on training data

Trained on 22K and tested on 4K frontal radiographs from a public dataset. Two pneumonia models were trained: (a) without and (b) with thoracic cavity annotations.

Model performance metrics

Area under the ROC curve (AUC) for detection of pneumonia was 0.861 and improved to 0.906 with concurrent training with the thoracic cavity annotation.

Performance Curves

curve1

curve2

Pneumonia detection ROC curve on public dataset od 3985 frontal Chest Xrays.

See reference publication for more information.

Model History

See more ▼less ▲
 
2020-3-5

Model available on Arterys.

AiDxLablogo

AiDA Lab at UC San Diego

AiDx™ is a mark of The Regents of the University of California.
The University California, San Diego is one of the world's leading public research universities, located in beautiful La Jolla, California

Brian Hurt

Brian Hurt

Learn more