Chest-X-Ray-Abnormality-Classification

This projects aims to detect abnormalities in chest X-rays using methods in Deep Learning.

Tasks

• Create a Dataset parser
  • Visualize images
  • Explore class imbalance
  • Process labels (one-hot encoding, reassigning labels, etc.)
• Binary classification
  • Create generators for train and validation splits
  • Train using a VGG-style neural network
  • Use transfer learning (InceptionResNetv2)
  • Plot training graphs (Accuracy, Loss)
• Multilabel classification
  • Create generators for train and validation splits
  • Without sampling: Train using a MobileNet pretrained on ImageNet
  • With sampling: Train using a MobileNet pretrained on ImageNet
  • Plot training graphs (Accuracy, Loss, ROC-AUC)

Technical specifications

The notebook is written in Python3 and uses the following major libraries/frameworks:

1. Deep Learning framework: Keras
2. Data processing: Numpy, Pandas, Scikit-learn
3. Data visualization: Matplotlib, OpenCV

Source

The data has been taken from the NIH Chest X-ray Dataset on Kaggle.

References

1. Monk AI's example notebook on multiclass classification using Satellite images
2. Scott Mader's notebook on multilabel classification of chest x-rays
3. Paul Mooney's notebook on predicting pathologies in X-Ray images