Efficient Tuberculosis Detection Using Chest X-ray Images with Deep Learning Algorithms

Abstract

Tuberculosis is a threat to the existence of the human race due to its substantial mortality rate and it has become a
significant public health concern, which if detected at an early stage could reduce the death rate globally.
Harnessing the potential of machine learning to combat the low detection rate of tuberculosis detection by traditional
methods and promote a faster and more accurate diagnosis of the disease. An online Dataset comprising 11,200
Chest X-ray (CXR) images of different categories of patients that are healthy, Sick but not infected, and those
infected with Tuberculosis with their corresponding bounding box annotations were used for this research, and
feature engineering was carried out on the dataset for effective data cleaning, Image resizing, normalization, and
augmentation to increase the quality of data during data segmentation. The dataset was divided into training,
validation, and testing sets using the RestNet50 model which demonstrated a good performance in classification,
achieving impressive precision, and recall of 98%, and 96% respectively, and YOLOv8 was also used with 68%
precision, 65% recall, and 68% mean average precision respectively which showed that the model needs
improvement to further accurately detect regions infected with tuberculosis.