Objectives To develop a CT radiomics model for predicting fibrosis grade in pancreatic ductal adenocarcinoma (PDAC) and to investigate the underlying prognosis value and biological basis. Methods Patients with resected PDAC were retrospectively included from three institutions. Evaluating tumor fibrosis content using fibrotic pixels proportion through Masson staining of postoperative pathological sections. Radiomics features from preoperative contrast-enhanced CT (CECT) were extracted and used to develop models in the training cohort. The diagnosis performance was further validated in the two test cohorts. The outcome cohort, including patients with advanced PDAC undergoing neoadjuvant chemotherapy, was used to evaluate the predictive value of the model for overall survival (OS) and disease-free survival (DFS), which were investigated using the Kaplan–Meier method and log-rank test. RNA sequencing data from a prospective biological basis cohort were conducted to explore the biological processes underlying the radiomics model. Results Among 215 patients (median age 60.89 years, 142 men) used for radiomics modeling, 132 (61.40%) were confirmed as high fibrosis content. The combined phase (CP) radiomics model, which included all CECT radiomics features, showed the best performance for predicting fibrosis grade, with AUCs of 0.831, 0.785, and 0.746 in training, internal test, and external test cohorts. OS (p = 0.011) and DFS (p = 0.022) can be categorized using the CP radiomics model in the outcome cohort. RNA-seq indicated that different CP models were associated with fibrotic production and remodeling processes. Conclusion The CP radiomics model showed the best performance in predicting fibrosis grades in PDAC. Critical relevance statement Fibrosis grading is of prognostic and neoadjuvant chemotherapy efficacy evaluation significance, and the CT-based combined phase radiomics model established in our study will facilitate risk stratification and selection of personalized treatment strategies for patients. Furthermore, underlying biological processes demonstrated in the radiomics model will offer valuable insights into their interpretability and clinical translation. Key Points <UnorderedList Mark="Bullet"> <ItemContent> Fibrosis grading is of prognostic significance in pancreatic ductal adenocarcinoma (PDAC), but lacks a reliable preoperative assessment. </ItemContent> <ItemContent> The CT-based combined phase (CP) radiomics model predicts fibrosis grading effectively in PDAC. </ItemContent> <ItemContent> The CP radiomics model demonstrated prognostic and neoadjuvant chemotherapy efficacy evaluation value and underlying biological processes, which related fibrotic production and remodeling processes. </ItemContent> </UnorderedList> Graphical Abstract

CT-based radiomics models decode fibrosis content and molecular differences in pancreatic ductal adenocarcinoma: a multi-institutional study

期刊论文
出版模式：

开放获取
发表日期：

2025年9月12日

查看全文

Fangqing Wang,
Yang Sun,
Jianwei Xu,
Yufan Chen,
Hui Zhang,
Guotao Yin,
Dexin Yu

摘要

Objectives

To develop a CT radiomics model for predicting fibrosis grade in pancreatic ductal adenocarcinoma (PDAC) and to investigate the underlying prognosis value and biological basis.

Methods

Patients with resected PDAC were retrospectively included from three institutions. Evaluating tumor fibrosis content using fibrotic pixels proportion through Masson staining of postoperative pathological sections. Radiomics features from preoperative contrast-enhanced CT (CECT) were extracted and used to develop models in the training cohort. The diagnosis performance was further validated in the two test cohorts. The outcome cohort, including patients with advanced PDAC undergoing neoadjuvant chemotherapy, was used to evaluate the predictive value of the model for overall survival (OS) and disease-free survival (DFS), which were investigated using the Kaplan–Meier method and log-rank test. RNA sequencing data from a prospective biological basis cohort were conducted to explore the biological processes underlying the radiomics model.

Results

Among 215 patients (median age 60.89 years, 142 men) used for radiomics modeling, 132 (61.40%) were confirmed as high fibrosis content. The combined phase (CP) radiomics model, which included all CECT radiomics features, showed the best performance for predicting fibrosis grade, with AUCs of 0.831, 0.785, and 0.746 in training, internal test, and external test cohorts. OS (p = 0.011) and DFS (p = 0.022) can be categorized using the CP radiomics model in the outcome cohort. RNA-seq indicated that different CP models were associated with fibrotic production and remodeling processes.

Conclusion

The CP radiomics model showed the best performance in predicting fibrosis grades in PDAC.

Critical relevance statement

Fibrosis grading is of prognostic and neoadjuvant chemotherapy efficacy evaluation significance, and the CT-based combined phase radiomics model established in our study will facilitate risk stratification and selection of personalized treatment strategies for patients. Furthermore, underlying biological processes demonstrated in the radiomics model will offer valuable insights into their interpretability and clinical translation.

Key Points

Fibrosis grading is of prognostic significance in pancreatic ductal adenocarcinoma (PDAC), but lacks a reliable preoperative assessment.

The CT-based combined phase (CP) radiomics model predicts fibrosis grading effectively in PDAC.

The CP radiomics model demonstrated prognostic and neoadjuvant chemotherapy efficacy evaluation value and underlying biological processes, which related fibrotic production and remodeling processes.

Graphical Abstract