Deep Learning for Chest Radiographs

This demo provides selectable models for:

  1. Heart and lung segmentation, with age, view, and sex prediction
  2. Pneumonia classification and segmentation
  3. Pneumothorax classification and segmentation
  4. Lines and tubes detection, segmentation, and placement classification

The segmentation heatmaps are based on pixel-level model outputs rather than non-explicit localization methods such as GradCAM.

This demonstration has not been approved by any regulatory agency. Do not upload protected health information.

Created by: Ian Pan, https://ianpan.me

Last updated: July 29, 2026
Models to run
Lines/tubes ensemble profile
1 5
0 5
0 5
0 5
Examples: pneumonia, pneumothorax, and held-out test images with abnormal ETT, NGT, or CVC

Example sources: The pneumonia example is my own personal chest radiograph. Pneumothorax case courtesy of Dr Ian Bickle, Radiopaedia.org, rID 74856 (CC BY-NC-SA 3.0). The ETT, NGT, and CVC examples are from the RANZCR CLiP test set.

Results