gerardmaggiolino / chestx-ray14-cnn

CNN architectures, training procedures, and evaluations for prediction of diseases on the ChestX-ray14 dataset.

Convolutional Neural Networks for Disease Classification

Three PyTorch implemented CNN architectures and their respective trainers for evaluation over the ChestX-ray14 dataset of thorax diseases. This dataset presents a class imbalanced multi-label classification problem, with 112,120 frontal-view X-rays labelled for fourteen different diseases. The architectures and trainers demonstrate use of the following techniques and optimizations:

• K-fold cross validation and early stopping with restoration of best models
• Undersampling of training set to address class imbalance problems
• Xavier initialization
• Confusion, Accuracy, Loss, Precision, and Recall recording and plotting
• Batch normalization
• Adam optimization
• ... and more! See modules for additional information.

Example training metrics and results

Training and Validation Set Performance

Class Confusion Matrices

CNN Filter Visualizations

Results of Class Balancing through Undersampling

Dataset citation:

Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2017, July). Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on (pp. 3462-3471). IEEE.