INVESTIGATING THE EFFECTIVENESS OF ADVANCED DEEP LEARNING MODELS IN PULMONARY DISEASE CLASSIFICATION USING CHEST X-RAY IMAGES

Abstract

According to the WHO's Global Tuberculosis Report 2022, an estimated 10.6 million people are infected with Tuberculosis and 1.6 million people die from Tuberculosis worldwide. Meanwhile, pneumonia is estimated to cause more than 700,000 deaths each year globally according to the UNICEF. In Vietnam, these two diseases are still serious health problems with a large number of cases and deaths. To support effective lung disease diagnosis, the study applied deep learning techniques, especially typical CNN models such as Resnet, DenseNet, Xception, MobileNet and InceptionV3 to classify chest X-ray images. These models are applied fine-tune transfer learning and then trained and compared to find the model with the highest accuracy. The results show that the MobileNet model is the best, with an accuracy of up to 98.31%, the highest compared to other models. The research results can be deployed and developed into a medical support system that helps doctors and medical staff quickly and accurately identify lung diseases from X-ray images, thereby helping to improve the diagnosis and treatment process for patients.