Radiomics and AI Advance NSCLC Staging Precision
A recent study published in Cancers explores the potential of artificial intelligence (AI) to enhance the preliminary staging of non-small cell lung cancer (NSCLC) using computed tomography (CT) imaging. Given that NSCLC remains one of the most prevalent and lethal cancers worldwide, timely and accurate staging is crucial for guiding treatment decisions and improving patient outcomes. Traditionally, staging relies on histological evaluation of biopsy or surgical samples, a process that can be time-consuming. The study aimed to develop an AI-based predictive model using a feed-forward neural network (FFNN) to classify NSCLC stages I, II, and III based on radiomics extracted from CT images. The model demonstrated high accuracy in both internal and external validation cohorts, suggesting that AI-assisted staging could facilitate earlier diagnostic prioritization and streamline the pathway to definitive histological confirmation.
The study utilized 2 independent datasets: the NSCLC-Radiomics dataset from The Cancer Imaging Archive (TCIA) for model training and internal testing, and a separate dataset from a local hospital for external validation. Patient demographics and 107 radiomic features were extracted from CT images, with tumor volume delineations verified by clinical oncologists. The FFNN model achieved an accuracy of 88.84% in training, 76.67% in internal testing, and 74.52% in external validation. These results indicate that the AI model effectively distinguishes between different NSCLC stages with balanced sensitivity and specificity. Compared with traditional machine-learning classifiers, the FFNN model outperformed support vector machines, decision trees, and ensemble classifiers, particularly in distinguishing stage II disease, which is often more challenging to classify due to its transitional characteristics between early and advanced stages.
The findings highlight the potential clinical applications of AI-based staging tools in NSCLC. By integrating radiomic data with neural network models, this approach could serve as a rapid, noninvasive triage tool to identify patients requiring expedited biopsy or surgical evaluation. The lightweight nature of the FFNN model makes it suitable for deployment on general-purpose radiology workstations, enabling real-time predictions based on routinely acquired CT scans. The study underscores the growing role of AI in oncology, offering a complementary method to support clinical decision-making while reducing diagnostic delays.
“In this study, we utilized the radiomics retrieved from CT images and patient demographics to build neural networks for overall staging prediction for NSCLC patients,” the researchers concluded. “The performance of the models built was evaluated through sensitivity, specificity, ROC curve, and overall accuracy. The proposed FFNN model demonstrated the best distinguishing ability and achieved very good performance in overall staging prediction in both internal and external cohorts of NSCLC patients.”
Reference
Cheung EYW, Kwong VHY, Ng KCF, et al. Overall staging prediction for non-small cell lung cancer (NSCLC): a local pilot study with artificial neural network approach. Cancers (Basel). 2025;17(3):523. doi:10.3390/cancers17030523
