A Deep Learning Approach for the Classification of Tuberculosis and Pneumonia Using NIH Dataset *

This research proposes a deep learning approach for accurately classifying tuberculosis (TB) and pneumonia from the National Institutes of Health (NIH) Chest X-Ray (CXR) dataset. The proposed model combines convolutional neural networks (CNNs) and Residual networks with other machine-learning classifiers for better feature extraction and classification. Image data augmentation artificially increases the dataset size by generating altered versions of the images in the dataset. This technique is beneficial in reducing overfitting for training [1]. To balance the classes, SMOTE (Synthetic Minority Over-sampling Technique) algorithm, undersampling of the majority classes, Borderline SMOTE and ADASYN (Adaptive Synthetic Sampling) techniques on the models are employed. Interestingly, the results obtained using the Borderline SMOTE technique are more promising than those obtained using other techniques. The novelty of this research lies in the implementation of both data augmentation and class balance techniques for the classification of tuberculosis and pneumonia. The results demonstrate high accuracy in differentiating TB and pneumonia, which can aid in early detection and treatment of these diseases.