Deep Learning-based Real-Time Detection of Emphysema in Chest Radiographs

In 2019, chronic obstructive pulmonary disease (COPD) led to 3.23 million deaths on a global scale, positioning it as the third most prevalent cause of mortality. Emphysema and chronic bronchitis are termed the World Health Organization (WHO) uses to describe COPD. The typical depiction of emphysema involves the damage to the tiny air sacs located at the ends of the lung airways. Chronic bronchitis causes airway irritation, coughing, and phlegm. Deep learning was applied to multiple CNN architectures (AlexNet, ResNet 50, ResNet 101, and DenseNet 201) to extract features from chest X-rays. Chest x-rays were classified as having or not having emphysema using machine learning classifiers, including nearest neighbor classifiers (kNN), kernel approximation classifiers (SVM kernel and logistic regression), neural network classifiers (NN), decision tree classifiers, Naive Bayes classifiers, discriminant analysis, and support vector machines (SVM). Densenet201+ SVM achieved 93.1% total validation accuracy across all 7548 images with an AUC of 0.9781, and test accuracy is 100% with an AUC of 1.0 across all 1510 images. Therefore, DenseNet 201+ SVM is the highest of the four methods across all 7548 images.