Enhancing Staging Accuracy for NSCLC Using Artificial Neural Networks

By: JNCCN 360 Staff
Posted: Monday, May 19, 2025

A pilot study published in Cancers found that the feed-forward neural network (FFNN) AI model demonstrated distinguishing ability and overall accuracy in predicting staging for non-small cell lung cancer (NSCLC). According to Eva Cheung, MD, of the University of Hong Kong, and colleagues, the FFNN AI model may have clinical application to help establish preliminary staging, with the added benefit of being able to predict multiple stages using one model.

The study authors used the NSCLC-radiomics dataset from The Cancer Imaging Archive for training and testing the AI model, along with a cohort from a local hospital. The data included the patients’ sex, age at diagnosis, survival days after confirmed diagnosis, TNM (tumor, node, metastasis) staging, and overall staging. During testing, the FFNN model achieved accuracies of 88.84% in overall accuracies in validation, 76.67% in internal cohort testing, and 74.52% in external cohort testing.

Additionally, the AI is lightweight and can be operated on a general-purpose computer, such as a CT-viewing workstation, allowing oncologists or staff to have easy access. “Clinical staff can click the FFNN model software, load the radiomics, and the overall staging can be predicted using the model built in this study automatically,” explained the study authors. “The resultant overall staging can be used to establish preliminary staging, which aids clinical staff in triaging patients with reference to their medical condition. The preliminary staging provides additional information for radiologists to determine whether further testing is required… The waiting time for diagnosis confirmation for NSCLC patients can be minimized, which may potentially improve their overall prognosis and quality of life.”

The study authors also suggest larger, multicenter studies to validate these findings across more diverse patient populations, as well as integrating other imaging modalities in the future, such as PET/CT or MRI.

Disclosures: For full author disclosures, visit mdpi.com.

Cancers