ADAN-Driven Tuberculosis Diagnosis: A Deep Learning Approach for Chest X-Rays

Recent advancements in medical imaging and deep learning have shown promise in the accurate identification of lung abnormalities, particularly Tuberculosis, through chest radiography. This paper proposed a new classification method for distinguishing between Tuberculosis and Normal chest X-rays, addressing imbalanced data challenges using the Adaptive Data Augmentation Network (ADAN) strategy during training. Using the Tuberculosis Chest X-ray dataset, the study used a modified ResNet50V2 feature extractor, followed by global pooling and multilayer perception techniques for classification. Through intelligent hyperparameter tuning facilitated by ADAN, the proposed method achieved a maximum accuracy of 0.98, with a recall of 1.00 for the normal class and 0.91 for the tuberculosis class. Furthermore, the study identified the impacting factors of the method based on obtained results, highlighting the significance of ADAN, feature extraction, and classification phases. The proposed method exhibits promising results, aiming to improve the detection rate of tuberculosis and expedite the diagnostic process.