Swin Transformer for Early Detection and Classification of Lung Cancer in Histopathological Images
Liharika Kola, Bhanu Prasad Rambha, Srisaila Ampalam, Ashok Kumar Munnangi · 2025
Lung cancer is one of the most common fatal cancers, that emphasizes the necessity of accurate and prompt detection techniques. The proposed approach utilizes deep learning to automatically detect and differentiate adenocarcinoma and squamous cell carcinoma from histological images. In a bid to capture key morphological features effectively in tissue samples, a Swin Transformer model has been utilized. With the Kaggle public-accessible Lung Cancer Histopathological Images dataset, the model is trained and validated. The model performed exceedingly well, with 99.8 % accuracy, 99.8 % precision, and 99.7 % recall. In contrast to the conventional convolutional neural networks, the transformer-based architecture enables better feature extraction, particularly when it comes to recognizing very small cellular and structural differences between LUAD and LUSC subtypes. These results indicate there is significant clinical potential for aiding pathologists in diagnosing lung cancer effectively and accurately. In support of future computational pathology research, the entire implementation and trained models are made freely available.