Enhancing Lung Cancer detection using Swin Vision Transformers

The uncontrolled growth of abnormal cells in the lungs causes tumors, which in turn affects the lung’s ability to function properly. While smoking is the leading cause, exposure to toxins, genetics, and pollution also contribute. Early detection is vital as advanced lung cancer spreads to other organs, complicating treatment and lowering survival rates. Imaging procedures start with chest X-rays and CT scans until healthcare providers perform biopsies for diagnosis and treatment stage determination. Traditional methods such as CNNs and Vision Transformers classify images into categories like normal, adenocarcinoma, and squamous cell carcinoma. However, they require high computational resources and often overlook small-cell lung cancer. To address these challenges, we are introducing Swin Vision Transformers for lung cancer detection. This approach stands out because it uses a shifted window mechanism that reduces computational costs while still capturing important features in medical images. Swin Transformers are also highly scalable and can handle images of different resolutions, making them well-suited for real-world applications. Our proposed model achieves a detection accuracy of $99.8 \%$, with a precision of $99 \%$ and recall of $99.7 \%$, significantly outperforming traditional CNN and ViT-based approaches.