Analyze Tuberculosis Using EfficientNet-B0 From Multi-Class Chest X-Ray

Computer vision is an emerging field in medical science, particularly in medical image classification. Deep learning (DL) has successfully classified chest conditions using X-ray (CXR) images. Most studies have traditionally focused on binary classifications, such as distinguishing between COVID-19 and healthy lungs or Tuberculosis (TB) and healthy lungs. This study extends previous efforts by utilizing the pre-trained and lightweight EfficientNet-B0 model to classify both binary and multi-class datasets. The multi-class dataset was created using images from the Mendeley CXR dataset, which contains images of COVID-19, Pneumonia, and healthy lungs, as well as the Mendeley TB dataset. The model is designed to classify four classes: COVID-19, Pneumonia, healthy lungs, and TB. The multi-class model achieved an accuracy of 99.57% and an F1 score of 99.59% on an unseen test sample from the combined Mendeley dataset, outperforming past literature on these four classes of CXR images. Additionally, a binary version of the model, focusing on distinguishing TB from non-TB (COVID-19, Pneumonia, and healthy lungs), achieved a perfect accuracy of 100% across all performance metrics.