The value of CT based radiomics in predicting progression of early acute pancreatitis
10.3760/cma.j.cn112149-20210829-00615
- VernacularTitle:CT影像组学预测早期急性胰腺炎进展的价值
- Author:
Haiyun FAN
1
;
Jiming CHEN
;
Liangliang CHEN
;
Lili WU
;
Hui ZHOU
Author Information
1. 皖南医学院弋矶山医院影像中心，芜湖 241001
- Keywords:
Pancreatitis, acute necrotizing;
Tomography, X-ray computed;
Radiomics
- From:
Chinese Journal of Radiology
2022;56(7):778-784
- CountryChina
- Language:Chinese
-
Abstract:
Objection To investigate the value of CT based radiomics in predicting progression of early acute pancreatitis (AP). Methods:From November 2013 to February 2021, 109 patients diagnosed with AP according to the new revised Atlanta classification in Yijishan Hospital of Wannan Medical College were retrospectively analyzed. The patients were divided into progressive group (40 cases) and non-progressive group (69 cases) according to the follow-up results. All patients underwent plain and enhanced abdominal CT scan within a week of onset. The patients were divided into training set (77 cases, including 28 cases in progressive group and 49 cases in non-progressive group) and validation set (32 cases, including 12 cases in progressive group and 20 cases in non-progressive group) in a ratio of 7∶3 using a computer completely random method. Manual region of interest mapping was performed on all levels of pancreas on the plain scan, arterial phase, venous phase and delayed phase CT images, then performed 3D fusion. AK software was used to extract texture features. The minimum redundancy maximum relevance and minimum absolute shrinkage and selection operator regression analysis were used to select features and establish radiomics labels of the plain scan, arterial phase, venous phase, delayed phase and combining the 4 sequences. The multiple logistic regression analysis was used to establish the clinical model by combining clinical features and CT features, and the comprehensive model was established by combining clinical features, CT features and imaging radiomics label. The receiver operating characteristic (ROC) curve was used to evaluate the efficacy of each model in predicting early AP progression and the decision curve analysis (DCA) was used to evaluate the clinical application value of each model.Results:In the training set, logistic regression results showed that edge was an independent predictor (OR=0.16, P=0.033). The clinical model was established using edge and serum calcium level, and its areas under the ROC curve (AUC) in the training set and validation set were 0.69 and 0.70, respectively. Totally 14, 11, 13, 13 and 9 optimal texture features were extracted from the plain scan, arterial phase, venous phase, delayed phase and combined sequence images, respectively. The delay phase image radiomics label had relatively better predictive performance in training set and validation set, and the AUC were both 0.85. The comprehensive model was established based on the delayed phase image radiomics label (OR=2.22, P<0.001) and edge (OR=0.02, P=0.042), and the AUC in the training set and validation set was 0.90 and 0.86, respectively. DCA showed that both the comprehensive model and the delayed phase radiomics label had better benefits. Conclusions:CT radiomics model has high value for predicting progression of AP, and its clinical benefits exhibited superior performance of clinical model.
