Efficient CT Image based Lung Cancer Detection Using Hybrid MobileViT–SAM-lite

Abstract

Early lung cancer is still a serious issue in medical imaging because the tumor structures are difficult to detect and the diagnosis is required as soon as possible. One of the major causes of cancer related deaths in the globe is lung cancer that has a huge percentage of cancer deaths and therefore there is a need to have intelligent computer aided diagnostic systems. More recent developments in deep learning have shown encouraging results in analyzing medical images but conventional convolutional neural networks can have difficulties capturing long-range contextual data whereas transformer-based architectures can add computational complexity. To overcome these shortcomings, the proposed study is to develop a hybrid deep learning model, which combines MobileViT, a lightweight Vision Transformer (ViT) that can use convolutional efficiency in addition to global attention mechanisms with Segment Anything Model (SAM) lite, an efficient segmentation engine to localize tumor boundaries. The novelty of the model is that it incorporates lightweight transformer-based feature extraction and segmentation-guided classification, which contributes to improved interpretability and less consumption of resources.The proposed model jointly segments and classifies lung cancer based on the CT images, allowing better representation of features and a high-level of diagnostic accuracy. The model attained a classification accuracy of 99.7%, indicating that it is a strong model in benchmarking benign, malignant, and normal cases.