A Comparative Study of Deep Learning Model for Assessing the Effectiveness for Classification of Healthy, Sick and Tuberculosis

The early diagnosis of tuberculosis (TB) is a significant global concern, with the World Health Organization (WHO) advocating for the incorporation of improved diagnostic techniques, such as high-quality chest X-rays (CXR) and laboratory assays, to address the TB detection gap. This work evaluates the efficacy of two advanced deep learning models-ResNet-50 and EfficientNet-B7-for classifying chest X-ray pictures into healthy, diseased, and tuberculosis categories. Utilizing transfer learning, pre-trained networks were refined on a labeled dataset to enhance feature extraction and augment classification accuracy. ResNet-50's architecture, characterized by its residual learning framework, promotes efficient gradient flow during training, hence improving its capacity to identify intricate patterns in CXR images. EfficientNet-B7, recognized for its compound scaling of depth, width, and resolution, presents a novel method for enhancing model performance while utilizing fewer parameters. Both models were assessed on accuracy, precision, recall, and F1-score to guarantee a thorough evaluation of their categorization abilities. The experimental findings indicated that ResNet-50 attained a flawless accuracy of 1.00, markedly surpassing EfficientNet-B7. The model frequently achieved elevated F1-scores, highlighting its exceptional equilibrium between precision and recall. The importance of these findings is attributed to ResNet-50's capability to accurately differentiate among normal, abnormal, and tuberculosis patients, rendering it exceptionally appropriate for tuberculosis detection in clinical environments. This study emphasizes the efficacy of deep learning models, specifically ResNet-50, in facilitating expedited and precise tuberculosis diagnosis by automated chest X-ray image analysis, hence contributing to the early identification and management of tuberculosis.