Automated Tuberculosis Detection in Chest Radiographs: A Hybrid Deep Learning Framework for Clinical Decision Support

Abstract

Tuberculosis (TB) remains a leading cause of death worldwide, particularly in resource-limited settings, where early detection is critical for controlling its spread. This study proposes a hybrid deep learning framework for the automated detection of TB from chest X-ray (CXR) images, integrating the strengths of three widely used Convolutional Neural Networks (CNNs): VGG16, ResNet50, and DenseNet121. The primary objective is to enhance the accuracy and reliability of TB diagnosis by combining the feature extraction capabilities of these models through a feature-level fusion strategy. The model was trained and evaluated on three publicly available datasets: the Montgomery County Chest X-ray dataset (138 CXR images), the Shenzhen Chest X-ray dataset (662 CXR images), and a third dataset containing 700 TB-positive and 3500 normal CXR images. Key performance metrics including accuracy, precision, recall, and F1-score were used to assess model performance. The hybrid model achieved a significant accuracy of 97.40%, with a precision of 0.96, recall of 0.96, and F1-score of 0.96, outperforming individual models such as VGG16 (93.00%), ResNet50 (95.90%), and DenseNet121 (96.30%). In comparison to existing methods, such as traditional manual interpretation and standalone deep learning models, the proposed hybrid model provides a more accurate, scalable, and robust solution for TB detection. The integration of complementary CNN architectures results in superior generalization, particularly in the face of dataset imbalances and varying image quality. This work demonstrates the potential of hybrid deep learning models to improve clinical decision support systems, offering a promising tool for rapid TB diagnosis, especially in resource-constrained healthcare settings.