LungPanelNet: a machine learning-based approach for the early prediction and differentiation of non-small cell lung...

Non-small cell lung cancer (NSCLC) represents a major global health challenge, primarily due to its frequent diagnosis at advanced stages, which significantly limits therapeutic efficacy and results in poor survival outcomes. A critical unmet need exists for non-invasive, accurate diagnostic tools for early detection.

This study aimed to develop and validate a robust machine learning model based on a panel of serum tumor markers for the early prediction of NSCLC and its differentiation from benign pulmonary conditions.

In this retrospective cohort study, we recruited 2,283 participants, including 1,339 with NSCLC, 313 with pneumonia, 260 with biopsy-confirmed benign lesions, and 371 with other benign lung masses. Serum levels of six key tumor markers-Squamous Cell Carcinoma Antigen (SCCA), Carcinoembryonic Antigen (CEA), Cancer Antigen 125 (CA-125), Cytokeratin 19 Fragment (CYFRA21-1), Neuron-Specific Enolase (NSE), and Pro-Gastrin-Releasing Peptide (ProGRP)-were quantified, and a custom deep neural network, LungPanelNet, was constructed for the classification task.

The model demonstrated superior predictive performance on an independent testing set, achieving an area under the receiver operating characteristic curve (AUC-ROC) of 0.92 (95% CI: 0.88-0.96), with an accuracy of 89.3%, a sensitivity of 91.5%, and a specificity of 87.8%. Feature importance analysis identified SCCA and CYFRA21-1 as the most significant predictors.

Our findings demonstrate that a machine learning model integrating a panel of serum tumor markers can effectively distinguish NSCLC from a spectrum of benign pulmonary conditions with high accuracy. This approach shows promise as a clinical decision-support tool, though further validation in larger, prospective, multi-center cohorts is warranted. This was a retrospective cohort study without clinical trial registration.