A Reproducible Multicentre MRI Radiomics Workflow for Pancreatic Cyst Risk Stratification Using Paired T1- and...

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Data Science

Background:

• Pancreatic cysts require accurate risk stratification to differentiate between benign and malignant lesions.
• Magnetic Resonance Imaging (MRI) offers detailed visualization of pancreatic cysts.
• Radiomics, the extraction of quantitative features from medical images, holds potential for improving risk assessment.

Purpose of the Study:

• To develop and technically validate a reproducible multicentre MRI radiomics workflow for pancreatic cyst risk stratification.
• To utilize paired T1- and T2-weighted imaging from public datasets for workflow development.
• To assess the performance of radiomics models in classifying pancreatic cyst risk.

Main Methods:

• Selection and processing of public pancreatic MRI datasets (Cyst-X cohort).
• Implementation of a quality-controlled radiomics workflow including resampling, intensity normalization, and feature extraction (PyRadiomics).
• Development and validation of predictive models using logistic regression and random forest classifiers on T2-weighted imaging with clinical data and paired T1/T2 data.

Main Results:

• A final cohort of 409 patients with 818 image-mask pairs was analyzed.
• The T2 + clinical logistic regression model achieved the highest macro-AUC of 0.737.
• The paired T1/T2 complete-case random forest model showed comparable performance (macro-AUC 0.735).

Conclusions:

• A reproducible public data MRI radiomics workflow for pancreatic cyst risk stratification is feasible.
• Current model performance is modest, indicating a need for further improvement.
• Independent external validation is essential before clinical implementation of this workflow.