ChestXRay

Chest X-Ray Pneumonia Classifiers Using fastai

Key Notebooks:

• Classification of Pneumonia/Normal

  • File: 'RESNET101 Classify Pneumonia.ipynb' GitHub nbviewer
  • Results Summary:
    • 94.0% accuracy tuning head only
    • 98.2% accuracy after fine tuning
    • 99.2% accuracy, 3.3% unknown with using threshold=0.80

• Classification of Viral/Bacterial Pneumonia

  • File: 'RESNET50 Classify Viral or Bacterial.ipynb' GitHub nbviewer
  • Results Summary:
    • 76.6% accuracy tuning head only
    • 77.8% accuracy after fine tuning
    • 82.7% accuracy, 25.26% Unknown with threshold = 0.70
    • 89.7% accuracy, 50.45% Unknown with threshold = 0.85

Kaggle Pneumonia Chest X-Ray Images Dataset

• dataset
• paper

Other Files of interest

• fastai_addons.py -- Various analysis functions for ClassificationInterpretation

Credits

• The basic Pneumonia Classifer notebook was modeled after https://www.kaggle.com/natevegh/pneumonia-detection-98-acc-fastai-2019-update