RaFiST: a radiomics model for non-invasive stratification of tumor fibrosis and prognostic prediction in non-small cell lung cancer.

objectivesTumor fibrosis plays a critical role in driving therapeutic heterogeneity and drug resistance. However, relevant research in non-small cell lung cancer (NSCLC) remains limited. This study aimed to determine the prognostic value of tumor fibrosis and develop a novel radiomics fibrosis stratification tool (RaFiST) for non-invasive patient stratification. MATERIALS AND

methodsIn this multicenter retrospective study of 532 patients with resected NSCLC, tumor fibrosis was histopathologically quantified via collagen fraction. RaFiST was developed using pre-treatment contrast-enhanced CT scans from training and external test cohorts. Prognostic performance was subsequently compared among clinical, direct radiomics, and combined models. Transcriptomic analysis investigated the model's underlying molecular mechanisms.

resultsMultivariable Cox regression revealed that the fibrosis score was an independent risk factor for disease-free survival (DFS) and overall survival (OS) at both centers. An optimal cutoff of 11.12% stratified patients into high- and low-fibrosis groups. RaFiST demonstrated excellent performance in predicting tumor fibrosis, achieving an area under the curve (AUC) of 0.879 in the training cohort and 0.813 in the test cohort. RaFiST-High patients exhibited significantly worse survival across both cohorts (all p < 0.01). Furthermore, the combined Clinical-RaFiST model outperformed the baseline clinical and direct radiomics models, yielding a 5-year DFS AUC of 0.827. Transcriptomic analysis associated the RaFiST-High group with pathways for extracellular matrix remodeling, hypoxia, and immune suppression.

conclusionTumor fibrosis is an independent prognostic risk factor in NSCLC. RaFiST provides a robust, non-invasive imaging biomarker for tumor fibrosis stratification and prognostic prediction. KEY POINTS: Question Tumor fibrosis plays a critical role in tumor therapeutic heterogeneity and drug resistance. However, relevant research in non-small cell lung cancer (NSCLC) remains relatively limited. Findings Tumor fibrosis is an independent prognostic factor in NSCLC, and radiomics fibrosis stratification tool (RaFiST), a CT-based radiomics model provides a robust non-invasive biomarker for fibrosis stratification and outcome prediction. Clinical relevance RaFiST offers a non-invasive, reproducible framework to evaluate tumor fibrosis in NSCLC, thereby enabling the earlier identification of postoperative patients with poor prognoses for prompt adjuvant therapy.