基于首次发作急性胰腺炎患者增强CT图像构建影像组学模型用于预测急性胰腺炎的复发。

回顾性纳入2019年2月至2023年6月收治的135例急性胰腺炎患者，随机分为训练组95例，验证组40例。从动脉和静脉期增强CT图像中提取和筛选影像组学特征，并收集患者临床因素，使用多因素Logistic回归构建影像组学评分模型、临床模型和和联合模型。采用接受者操作特性曲线（ROC）、ROC曲线下面积（AUC）以及预测指标评价三组模型的预测能力，临床决策曲线分析（DCA）用于对比不同模型临床决策的净收益情况。将联合模型制作为列线图，提供预测急性胰腺炎患者复发风险概率。

训练组95例患者中有33例复发急性胰腺炎，验证组40例患者中有13例复发，分为复发组46例，非复发组89例。经影像组学特征降维及临床因素筛选，四种最优影像组学特征以及高脂血症和改良CT严重指数两种临床因素被分别用于并构建影像组学评分模型和临床模型。影像组学评分模型在训练组和验证组队列中表现出较好的预测效果，并优于临床模型（AUC_训练组：0.874比0.713，P=0.01；AUC_验证组：0.822比0.734，P=0.036）。联合模型可提高对胰腺炎复发预测的临床净获益率。

基于增强CT图像的放射组学评分模型在提高急性胰腺炎复发个体化预测的能力方面具有显著优势，有助于临床预防胰腺炎的复发和治疗。

To construct a radiomics-score model based on enhanced CT images of patients with first-time acute pancreatitis for the prediction of recurrence.

A total of 135 patients with acute pancreatitis treated at our hospital from February 2019 to June 2023 were included. The study cohort was randomly divided into a training group (n=95) and a validation group (n=40). Radiomics features were extracted and selected from arterial and venous phase enhanced CT images. Clinical factors were collected, and a multifactor logistic regression was used to build radiomics-score model, clinical models, and combined model. Receiver operating characteristic (ROC) curve, area under the ROC curve (AUC), and predictive indicators were employed to assess the predictive capabilities of the three models. Decision curve analysis (DCA) was used to compare the net benefits of different models in clinical decision-making. The combined model was presented as a nomogram to provide the risk probability of recurrence for patients with acute pancreatitis.

Thirty-three of 95 patients in the training group had recurrent acute pancreatitis, and 13 of 40 patients in the verification group had recurrent pancreatitis, which was divided into 46 patients in the recurrent group and 89 patients in the non-recurrent group. After dimensionality reduction of radiomics features and clinical factor selection, four optimal imageomics features and two clinical factors (hyperlipidemia and modified CT severity index) were used to construct radiomics-score model and clinical model, respectively. The radiomics-score model demonstrated better predictive performance in both the training and validation groups compared to the clinical model (AUC_{training group}: 0.874 vs. 0.713, P=0.01; AUC_{validation group}: 0.822 vs. 0.734, P=0.036). The combined model improved the clinical net benefit in predicting pancreatitis recurrence.

The radiomics-score model based on enhanced CT images shows significant advantages in enhancing individualized prediction of acute pancreatitis recurrence. This model can be valuable in clinical practice for the prevention and treatment of pancreatitis recurrence.