Frequency Ranking of Imaging Biomarkers for Lung Cancer Risk Stratification Using a Hybrid Elastic Net Method

TL;DR

This study shows that a CT scan-based imaging feature called Busyness can better predict lung cancer survival than traditional factors like age and tumor stage.

Contribution

The study introduces Busyness as a novel, noninvasive imaging biomarker for lung cancer prognosis that outperforms conventional clinical variables.

Findings

Busyness, a texture-based imaging feature, outperformed tumor stage, age, and sex in predicting survival outcomes in lung cancer patients.

Stratified analyses confirmed Busyness as a consistent predictor of survival across different age and sex groups.

The use of SMOTE improved model robustness and validated the stability of Busyness as a prognostic biomarker.

Abstract

Accurate prediction of survival in lung cancer remains challenging, as patients with similar clinical characteristics often experience remarkably different outcomes. Traditional prognostic indicators such as tumor stage, age, and sex do not fully reflect underlying tumor aggressiveness. Medical imaging, which is already part of routine clinical care, contains additional quantitative information that can be leveraged to improve risk assessment. In this study, we demonstrate that a texture-based imaging feature, called Busyness, extracted from standard CT scans, serves as a strong indicator of survival in patients with non-small-cell lung cancer. This imaging biomarker consistently distinguishes high- and low-risk patients more effectively than conventional clinical factors across different age and sex groups. These findings suggest that imaging-derived biomarkers can enhance clinical decision-making by providing a noninvasive and objective measure of tumor behavior, thereby supporting more personalized lung cancer management.

Lung cancer remains the leading cause of cancer-related mortality worldwide, emphasizing the critical need for novel and robust biomarkers to improve prognostication and guide precision oncology. While traditional clinical variables such as tumor stage, age, and sex are routinely used for survival prediction, their prognostic performance is limited. Imaging biomarkers derived from radiomic analysis of advanced medical imaging have emerged as a promising class of noninvasive cancer biomarkers, enabling quantitative characterization of tumor phenotypes. In this study, we investigated the prognostic utility of radiomic imaging biomarkers, with a particular focus on the texture-based feature Busyness, and compared their performance against conventional clinical factors. Survival analyses demonstrated that Busyness achieved significantly stronger discrimination of survival outcomes than stage, age, or sex. Stratified analyses further showed that Busyness consistently remained a dominant predictor of survival across age and sex subgroups, whereas tumor stage alone provided limited prognostic separation. To address class imbalance and enhance model robustness, the Synthetic Minority Over-sampling Technique (SMOTE) was applied, further supporting the stability of the imaging biomarker findings. These results highlight the potential of radiomic imaging biomarkers as powerful prognostic tools in lung cancer and support their integration into clinical workflows. This work contributes to the growing landscape of new cancer biomarkers and provides a foundation for future studies integrating imaging biomarkers with molecular and genomic markers to achieve improved prognostic accuracy.

Figures