Abstract

Introduction/Objective: Early identification of lung cancer can save lives and reduce mortality. Existing methods struggle to jointly capture local features, global relationships, and temporal changes, limiting reliable lung nodule detection and classification.

Methods: This study presents NoduleVision-Net, a hybrid deep learning framework that can aid in the future potential detection of lung cancer in its early and clinically relevant stages. Following the extraction of clean and consistent input data through the preprocessing steps of Non-Local Means filtering and brightness normalization, the framework merges convolutional-based modules that extract local features with transformer-based modules that understand a global context, while temporal modeling can understand patterns of nodule growth and development over time. The Nodule-Net framework creates a bridge between local and global representations and ultimately provides a novel, robust, and clinically interpretable approach for detecting and classifying lung nodules.

Results: The Nodule-Net model demonstrated superior performance in lung cancer classification, attaining 99.98 % accuracy, precision, recall, and F1-score on the Chest CT dataset. It surpasses existing approaches while enhancing predictive reliability and reducing computational demands.

Discussion: This proposed model demonstrates its robust performance in lung cancer detection by combining high accuracy with computational efficiency. Its high precision and reduced complexity outperform prior methods by enhancing early diagnosis, informing clinical decision-making, and advancing automated imaging analysis in medical practice.

Conclusion: This model, with high accuracy, precision, and reduced complexity in lung cancer classification, outperforms traditional methods by enhancing early diagnosis, informing clinical decision-making, and advancing automated imaging analysis in medical practice.

Statistics