From Pixels to Pathology: The Power of CNNs in Detecting Tuberculosis

INTRODUCTION: Tuberculosis (TB) remains a significant global health threat, demanding trustworthy and effective detection techniques. This study investigates the utilization of deep learning models, specifically ResNet50, InceptionV3, AlexNet, DenseNet121, and Inception3, for diagnosing tuberculosis from chest X-ray images. With a substantial dataset comprising 4,000 chest X-ray images, sourced from seven different nations and categorized as TB-infected or normal, this research aims to evaluate the performance of various deep learning architectures in accurately distinguishing TB instances. OBJECTIVES: The primary objective of this study is to assess the efficacy of different deep learning models in differentiating TB instances from chest X-ray images. By employing segmentation, data augmentation, and image pre-processing techniques, the research aims to enhance model performance and reliability in TB diagnosis. METHODS: The chest X-ray image dataset, scaled to 224x224 pixels, underwent segmentation, data augmentation, and pre-processing before being fed into the deep learning models. The dataset was divided into 80% for model training and 20% for testing, utilizing a five-fold cross-validation technique. Performance evaluation metrics including accuracy, precision, recall, and F1-score were employed to assess the models' effectiveness in TB identification. RESULTS: The findings indicate that ResNet50 and InceptionV3 models achieved near-perfect accuracy, precision, recall, and F1-scores, demonstrating their potential as reliable methods for TB identification. Despite exhibiting lower accuracy for the TB class, AlexNet also displayed good performance. However, DenseNet121 and Inception3 models showed room for improvement, particularly in precision and recall for the TB class. CONCLUSION: This study underscores the potential of deep learning models in enhancing TB identification in chest X-ray images. It highlights the importance of segmentation, data augmentation, and image pre-processing techniques in improving model performance. Future research may explore hyperparameter tuning, alternative data augmentation strategies, and ensemble approaches to optimize the performance of these models further. Overall, this work contributes to the growing body of knowledge on the application of artificial intelligence in healthcare, particularly in disease diagnosis and detection.