LungGuard: An IoT-Integrated Wireless Sensor and Hybrid CNN–Vision Transformer Framework for Early Lung Cancer Detection and Risk Assessment

Abstract

Lung cancer is a significant health problem worldwide, with timely diagnosis of lung abnormalities being crucial for enhancing diagnostic and treatment results. Interpretation of computed tomography (CT) images is difficult, however, because of the great variability in pulmonary nodules in size, shape, density, texture and anatomical setting. In this study, LungGuard, a CNN–Vision Transformer hybrid network for early lung-cancer detection, classification and risk assessment, is proposed. The proposed framework is based on LIDC-IDRI dataset and includes lung segmentation, nodule localization, region-of-interest extraction and CT preprocessing. A CNN is used to extract the local morphological and textural features, and a Vision Transformer is used to model the global context via self-attention. These complementary representations are then fused adaptively with feature fusion and used for multi-task prediction for nodule detection, malignancy classification, and risk assessment. The risk-assessment module can also include relevant clinical variables. To improve the interpretability of the model, grad-CAM and transformer attention visualization are added. The accuracy, precision, sensitivity, specificity, F1 Score, ROC-AUC, precision-recall AUC and calibration measures will be used to assess the performance. The proposed framework is designed to offer an understandable computer-aided approach to intelligent lung-cancer evaluation.