Lung cancer classification and severity estimation using vision transformer-IPCNN with enhanced U-Net segmentation

Introduction The death rate due to lung cancer is increasing because of late detection. Several automated lung cancer diagnostic techniques based on Machine Learning (ML) and Deep Learning (DL) have been developed for early detection. However, existing methods have limitations, including higher false-positive rates, lower sensitivity, and greater time consumption, which can affect early lung cancer diagnosis. Methods To address these limitations, the Lung Cancer Classification and Severity Estimation–Improved Parallel Convolutional Neural Network-Vision Transformer (LCCSE-IPCNN-ViT) model is proposed. The framework consists of preprocessing, segmentation, classification, and severity estimation phases. First, the input images are preprocessed using the Improved Wiener Filtering (IWF) technique. The preprocessed images are then segmented using the Modified TransUNet (MTransUNet) model. Finally, the segmented images are classified and their severity is estimated using the IPCNN-ViT model, which combines the Improved Parallel Convolutional Neural Network (IPCNN) and Vision Transformer (ViT). Results The proposed LCCSE-IPCNN-ViT model achieved an accuracy of 97.3%, sensitivity of 97.8%, specificity of 96.8%, precision of 96.8%, F-measure of 97.3%, and ROC-AUC of 98.5%. Discussion The experimental results demonstrate that the proposed LCCSE-IPCNN-ViT model provides accurate and efficient lung cancer classification and severity estimation, indicating its potential for supporting early lung cancer diagnosis.