- I made an image classifier to detect Pneumothorax (collapsed lung) on x-ray scans. It is trained on a Resnet 50 CNN model.
- AUC was recorded at 89% for the best version of the model. Pretty good compared with the 96% AUC of a CNN model made by Gooßen, et. al. and considering this was done within 2 days.
- Data was taken on a synthetic dataset provided by the National University Hospital (NUH) and Massachusets Institute of Technology Critical Care (MIT Critical Care). Data is not provided.
- Pneumothorax_code.ipynb: Main notebook I used to train the model. I retroactively added markdown for clarity.
- Test_1024.ipynb: Saved checkpoint of 1024 x 1024 images but I discontinued training because it was inefficient. Code is provided as-is.
-
Notifications
You must be signed in to change notification settings - Fork 1
ajsanjoaquin/Pneumothorax
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
Image classifier to detect Pneumothorax (collapsed lung) on x-ray scans. It is trained on a Resnet 50 CNN model.
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published