Towards Reliable Lung Cancer Diagnosis: A Novel CNN-RNN Hybrid Model For CT-Based Detection

Abstract

To improve the classification accuracy of lung cancer utilizing the LIDC and Chest CT datasets, this study presents the LungNet-RC model, a hybrid CNN-RNN architecture. With data pre-processing techniques like box filtering and contour enhancement, the model achieves exceptional accuracy, specificity, sensitivity, and AUC scores. Specifically, LungNet-RC attains overall accuracy as 0.9878, training accuracy as 0.9937, test accuracy as 0.9937 and validation accuracy as 0.87 on the LIDC dataset and 0.9988 overall accuracy on the Chest CT dataset with train accuracy as 0.9926, test accuracy as 0.9975 and validation accuracy as 0.7656. When compared with recent deep learning and fusion models, LungNet-RC’s superior performance demonstrates its reliability for clinical applications. This study’s findings underscore the robustness of the model in differentiating between malignant and benign cases, presenting LungNet-RC as a promising tool for early lung cancer detection.