Predictive model construction of anastomotic thickening character after radical surgery of esophageal cancer based on CT radiomics and its application value
10.3760/cma.j.cn115610-20230918-00106
- VernacularTitle:基于CT检查影像组学食管癌根治术后吻合口增厚性质预测模型的构建及其应用价值
- Author:
Jingjing XING
1
;
Yaru CHAI
;
Pengchao ZHAN
;
Fang WANG
;
Junqiang DONG
;
Peijie LYU
;
Jianbo GAO
Author Information
1. 郑州大学第一附属医院放射科，郑州 450052
- Keywords:
Esophageal neoplasms;
Squamous cell carcinoma;
Tomography, X-ray com-puted;
Radiomics;
Anastomotic stenosis;
Anastomotic recurrence
- From:
Chinese Journal of Digestive Surgery
2023;22(10):1233-1242
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the predictive model construction of anastomotic thickening character after radical surgery of esophageal cancer based on computed tomogralphy(CT) radiomics and its application value.Methods:The retrospective cohort study was conducted. The clinicopathological data of 202 patients with esophageal squamous cell carcinoma (ESCC) who were admitted to The First Affiliated Hospital of Zhengzhou University from January 2013 to June 2021 were collected. There were 147 males and 55 females, aged (63±8) years. Based on random number table, 202 patients were assigned into training dataset and validation dataset at a ratio of 7:3, including 141 cases and 61 cases respectively. Patients underwent radical resection of ESCC and enhanced CT examination. Observation indicators: (1) influencing factor analysis of malignant anas-tomotic thickening; (2) construction and evaluation of predictive model; (3) performance comparison of 3 predictive models. The normality of continuous variables was tested by Kolmogorov-Smirnov method. Measurement data with normal distribution were represented as Mean± SD, and comparison between groups was analyzed using the t test. Measurement data with skewed distribution were represented as M( Q1, Q3), and comparison between groups was analyzed using the Mann-Whintney U test. Count data were represented as absolute numbers, and comparison between groups was analyzed using the chi-square test or Fisher's exact probability. The consistency between subjective CT features by two doctors and measured CT numeric variables was analyzed by Kappa test and intraclass correlation coefficient (ICC), with Kappa >0.6 and ICC >0.6 as good consistency. Univariate analysis was conducted by corresponding statistic methods. Multivariate analysis was conducted by Logistics stepwise regression model. The receiver operating characteristic (ROC) curve was drawn, and area under curve (AUC), Delong test, decision curve were used to evaluate the diagnostic efficiency and clinical applicability of model. Results:(1) Influencing factor analysis of malignant anastomotic thickening. Of the 202 ESCC patients, 97 cases had malignant anastomotic thickening and 105 cases had inflammatory anastomotic thickening. The consistency between subjective CT features by two doctors and measured CT numeric variables showed Kappa and ICC values >0.6. Results of multivariate analysis showed that the maximum thickness of anastomosis and CT enhancement pattern were independent influencing factors for malignant anastomotic thickening[ hazard ratio=1.46, 3.09, 95% confidence interval ( CI) as 1.26-1.71,1.18-8.12, P<0.05]. (2) Construction and evaluation of predictive model. ① Clinical predictive model. The maximum thickness of anasto-mosis and CT enhancement pattern were used to construct a clinical predictive model. ROC curve of the clinical predictive model showed an AUC, accuracy, sensitivity, specificity as 0.86 (95% CI as 0.80-0.92),0.77, 0.77, 0.80 for the training dataset, and 0.78 (95% CI as 0.65-0.89), 0.77, 0.77, 0.80 for the validation dataset, respectively. Results of Delong test showed no significant difference in AUC between the training dataset and validation dataset ( Z=1.22, P>0.05). ② Radiomics predictive model. A total of 854 radiomics features were extracted and 2 radiomics features (wavelet-LL_first order_ Maximum and original_shape_VoxelVolume) were finally screened out to construct a radiomics predictive model. ROC curve of the radiomics predictive model showed an AUC, accuracy, sensitivity, specificity as 0.87 (95% CI as 0.81-0.93), 0.80, 0.75, 0.86 for the training dataset, and 0.73 (95% CI as 0.63-0.83), 0.80, 0.76, 0.94 for the validation dataset, respectively. Results of Delong test showed no significant difference in AUC between the training dataset and validation dataset ( Z=-0.25, P>0.05). ③ Combined predictive model. Results of multivariate analysis and radiomics features were used to construct a combined predictive model. ROC curve of the combined predictive model showed an AUC, accuracy, sensitivity, specificity as 0.93 (95% CI as 0.89-0.97),0.84, 0.90, 0.84 for the training dataset, and 0.79 (95% CI as 0.70-0.88), 0.89, 0.86, 0.91 for the validation dataset, respectively. Results of Delong test showed no significant difference in AUC between the training dataset and validation dataset ( Z=0.22, P>0.05). (3) Performance comparison of 3 predictive models. Results of Hosmer-Lemeshow goodness-of-fit test showed that the clinical predictive model, radiomics predictive model and combined predictive model had a good fitting degree ( χ2=4.88, 7.95, 4.85, P>0.05). Delong test showed a significant difference in AUC between the combined predictive model and clinical predictive model, also between the combined predictive model and radiomics predictive model ( Z=2.88, 2.51, P<0.05 ). There was no significant difference in AUC between the clinical predictive model and radiomics predictive model ( Z=-0.32, P>0.05). The calibration curve showed a good predictive performance in the combined predictive model. The decision curve showed a higher distinguishing performance for anastomotic thickening character in the combined predictive model than in the clinical predictive model or radiomics predictive model. Conclusions:The maximum thickness of anastomosis and CT enhancement pattern are independent influencing factors for malignant anastomotic thickening. Radiomics predictive model can distinguish the benign from malignant thickening of anastomosis. Combined predictive model has the best diagnostic efficacy.
