Enhancing Diagnostic Accuracy in Lung Disease:A VIT Approach

In the realm of medical diagnostics, where limited data availability poses a challenge, this project addresses the detection of pulmonary diseases, tuberculosis, Corona Virus using chest X-Radiation images from modest datasets containing fewer than a thousand samples. By harnessing the potential of deep Vision Transformer approach is employed for lung disease classification. A distinctive pipeline is established, encompassing image preceding classification. Comparative analysis with existing frameworks showcases the efficacy of this approach. Remarkably, the study reveals that pretrained models and even simpler classifiers like shallow neural networks can rival complex systems. The framework is also validated against publicly available lung datasets, positioning it as a competitive solution. This method achieves accuracy levels akin to state-of-the-art models while embracing the advantage of fewer trainable parameters. Importantly, the MobileViT based model showcases performance parity with superior solutions, effectively reconciling computational efficiency with diagnostic accuracy. This project pioneers an innovative pathway, transcending data constraints to empower accurate and accessible pulmonary disease detection through a resource-efficient framework.