Tomography (Ann Arbor, Mich.) · 2026
A Reproducible Multicentre MRI Radiomics Workflow for Pancreatic Cyst Risk Stratification Using Paired T1- and T2-Weighted Imaging.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Sgourakis George
Аннотация
PURPOSE: To develop and technically validate a reproducible multicentre MRI radiomics workflow for pancreatic cyst risk stratification using paired T1- and T2-weighted imaging from public datasets.
METHODS: Public datasets were screened and Cyst-X was selected as the primary cohort because it contained pancreatic MRI, risk labels, masks and metadata. A linked Cyst-X subset was enriched with metadata, filtered to an exact paired T1/T2 cohort, and processed through image-mask quality control, 1.0 mm isotropic resampling, intensity normalisation, PyRadiomics feature extraction, feature reduction and patient-level centre-held-out validation. The revised modelling strategy used a T2 + clinical all-patient primary analysis (n = 409) and a complete-case paired T1/T2 sensitivity analysis (n = 299).
RESULTS: The final cohort comprised 409 patients and 818 image-level rows across EMC, IU, MCF and NYU. All 818 image-mask pairs passed post-preprocessing QC. T2 radiomics were complete for all 409 patients; however, 110 T1 feature sets were missing, all from MCF. In the all-patient T2 + clinical model comparison, logistic regression achieved the highest macro-AUC (0.737). The T2 + clinical random forest comparator achieved macro-AUC 0.716 (95% CI 0.678-0.755), accuracy 0.545 (95% CI 0.496-0.592) and macro-F1 0.530 (95% CI 0.481-0.577). The paired T1/T2 complete-case random forest sensitivity model achieved macro-AUC 0.735 (95% CI 0.691-0.777), accuracy 0.575 (95% CI 0.520-0.632) and macro-F1 0.554 (95% CI 0.494-0.605).
CONCLUSION: This study demonstrates the feasibility of constructing a reproducible public data MRI radiomics workflow for pancreatic cyst risk stratification. Model performance was modest, and independent external validation is required before clinical application.
METHODS: Public datasets were screened and Cyst-X was selected as the primary cohort because it contained pancreatic MRI, risk labels, masks and metadata. A linked Cyst-X subset was enriched with metadata, filtered to an exact paired T1/T2 cohort, and processed through image-mask quality control, 1.0 mm isotropic resampling, intensity normalisation, PyRadiomics feature extraction, feature reduction and patient-level centre-held-out validation. The revised modelling strategy used a T2 + clinical all-patient primary analysis (n = 409) and a complete-case paired T1/T2 sensitivity analysis (n = 299).
RESULTS: The final cohort comprised 409 patients and 818 image-level rows across EMC, IU, MCF and NYU. All 818 image-mask pairs passed post-preprocessing QC. T2 radiomics were complete for all 409 patients; however, 110 T1 feature sets were missing, all from MCF. In the all-patient T2 + clinical model comparison, logistic regression achieved the highest macro-AUC (0.737). The T2 + clinical random forest comparator achieved macro-AUC 0.716 (95% CI 0.678-0.755), accuracy 0.545 (95% CI 0.496-0.592) and macro-F1 0.530 (95% CI 0.481-0.577). The paired T1/T2 complete-case random forest sensitivity model achieved macro-AUC 0.735 (95% CI 0.691-0.777), accuracy 0.575 (95% CI 0.520-0.632) and macro-F1 0.554 (95% CI 0.494-0.605).
CONCLUSION: This study demonstrates the feasibility of constructing a reproducible public data MRI radiomics workflow for pancreatic cyst risk stratification. Model performance was modest, and independent external validation is required before clinical application.