MRI T2加权影像组学联合临床变量预测高甘油三酯性急性胰腺炎病情严重程度
本研究回顾性纳入207例高甘油三酯性急性胰腺炎(HTG-AP)患者,基于T2加权MRI提取胰腺影像组学特征并结合临床变量构建随机森林模型,预测重症急性胰腺炎(SAP)。模型在训练集、内部测试集和外部验证集的AUC分别为0.955、0.954和0.930,显著优于仅影像组学、仅临床、BISAP或MRSI模型。该无创模型有望用于早期风险分层。
要点
- T2加权MRI影像组学联合临床变量可高精度预测HTG-AP进展为SAP。
- 模型外部验证AUC达0.930,优于BISAP和MRSI等临床评分。
- 胰腺手动勾画基于T2加权图像,使用3D Slicer和PyRadiomics提取特征。
局限性
- 回顾性研究设计
- 外部验证样本量较小(55例)
- 仅基于单家医疗集团数据
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
OBJECTIVES: Hypertriglyceridemic acute pancreatitis (HTG-AP) carries a high risk of severe disease (HTG-SAP). Early recognition of patients likely to progress to HTG-SAP is crucial for timely intervention. This study aimed to develop and validate a combined T2-weighted MRI (magnetic resonance imaging) radiomics-clinical model for accurate, noninvasive early prediction of SAP in HTG-AP and to compare its performance with established clinical scores.
METHODS: This retrospective analysis incorporated a derivation cohort of 207 patients with HTG-AP, who were classified as non-SAP (n=159) or SAP (n=48) based on the 2012 revised Atlanta criteria. These patients were randomly assigned to internal training and internal test sets at a 7:3 ratio. An additional 55 patients from an independent hospital campus within the same medical group were collected as an external validation cohort. The pancreas was manually delineated on T2-weighted images using 3D Slicer, while extraction of radiomic features was undertaken using PyRadiomics. After testing reproducibility and performing multistep feature selection, including LASSO regression, a random-forest model based on clinical and radiomic features was constructed. Predictive performance was evaluated using AUC values from ROC curves, DeLong's test, calibration curves, and decision curve analysis. For further robust assessment in the imbalanced dataset, Precision-Recall (PR) curves and their corresponding PR-AUC were also computed. The model, trained on the internal training set of the derivation cohort, was then applied to predict outcomes in the external validation cohort, with performance assessed using the same metrics.
RESULTS: The integrated radiomics-clinical model, which included six radiomic descriptors and four clinical indicators, yielded AUCs of 0.955 (95% CI: 0.914-0.997) in the training set and 0.954 (95% CI: 0.907-0.996) in the internal test set, and 0.930 (95% CI: 0.865-0.995) in the external validation cohort. These values were substantially higher than those obtained with radiomics-only, clinical-only, BISAP, or MRSI models. Additionally, the model demonstrated robust performance on PR analysis, with PR-AUCs of 0.906 (95% CI: 0.820-0.971) in the internal training set and 0.868 (95% CI: 0.685-0.977) in the internal test set. The model also demonstrated good calibration and achieved the highest net benefit over a wide range of decision thresholds.
CONCLUSIONS: A model combining MRI T2-weighted radiomics with clinical variables enables highly accurate early prediction of SAP in HTG-AP. This noninvasive approach holds promise for facilitating early risk stratification and supporting personalized treatment strategies in clinical practice.
引用
- Li Y, Wang Y, Wan X, et al. Radiomics derived from MRI T2-weighted imaging combined with clinical variables for predicting disease severity in hypertriglyceridemic acute pancreatitis. Abdom Radiol (NY). 2026. PMID: 41925844. doi:10.1007/s00261-026-05487-0
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。