Deep Learning with Transformer-Based Feature Extraction for Enhanced Lung Nodule Detection and Classification In Medical Imaging

Abstract

Lung cancer continues to be a major contributor to cancer-related fatalities worldwide, emphasizing the critical need for early and precise diagnosis to enhance patient survival rates. Recent progress in deep learning, particularly in transformer-based models, has significantly transformed medical image analysis by utilizing self-attention mechanisms to extract complex spatial and contextual features. This research introduces an advanced deep transformer-driven framework for feature extraction, designed to improve lung nodule detection and classification in computed tomography (CT) scans. Unlike traditional convolutional neural networks (CNNs), which are constrained by local receptive fields, transformers effectively capture long-range dependencies in medical images, enabling the recognition of subtle textural and morphological patterns associated with malignancies. By incorporating multi-scale feature representations and self-attention modules, the proposed model enhances the distinction between benign and malignant nodules while reducing false-positive occurrences. Comprehensive evaluations on established CT datasets reveal superior performance compared to conventional deep learning methods, demonstrating the potential of transformer-based architectures in lung nodule detection. These findings underscore the transformative role of transformers in automated medical imaging, contributing to the development of more interpretable and generalizable deep learning frameworks for clinical applications.