Automated Pulmonary Tuberculosis Detection in Chest Radiographs using Pretrained DCNN Models

The problem of pulmonary tuberculosis (PTB) worldwide is still considered to be a significant one, and it is therefore necessary to develop precise and effective diagnostic tools that will make it possible to carry out early interventions and treatment. In this research, the deep convolutional neural network (DCNN) architectures named VGG-16, VGG-19, ResNet-50, ResNet-101, and MobileNet were utilized for the detection of PTB from chest radiographs. By means of transfer learning and fine-tuning approaches it was possible to increase the diagnostic performance. Notably, VGG-16 consistently demonstrated exceptional performance, achieving a remarkable accuracy of 99.524% in both transfer learning and fine-tuning phases. Similarly, MobileNet exhibited strong performance, with an accuracy of 99.524% in transfer learning and 98.095% in fine-tuning. The results obtained show that VGG-16 and MobileNet are very good for the detection of PTB from chest X-ray images. Moreover, the fine-tuning showed the best performance of ResNet models, therefore, it confirmed the effectiveness of the iterative refinement process. In general, the results mainly highlight the significance of transfer learning and fine-tuning in the best model parameter adjustment, with VGG-16 being the most effective model. Possible future directions will be investigated through ensemble methods, domain-specific knowledge integration, advanced augmentation techniques, and multimodal data sources that will aim to improve PTB detection accuracy and deal with the changing healthcare challenges.