Development of a random forest prediction model for perineural invasion in gallbladder carcinoma based on preoperative enhanced CT image features
10.3760/cma.j.cn113884-20240322-00083
- VernacularTitle:基于术前增强CT影像特征的胆囊癌神经浸润随机森林预测模型的构建
- Author:
Min YANG
1
;
Qi LI
;
Wenli HUO
;
Wenzhi LI
;
Na LI
;
Zhimin GENG
;
Jian YANG
Author Information
1. 西安交通大学第一附属医院医学影像科，西安 710061
- Keywords:
Gallbladder neoplasms;
Enhanced computed tomography;
Perineural invasion;
Risk factor;
Random forest
- From:
Chinese Journal of Hepatobiliary Surgery
2024;30(8):581-585
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a random forest prediction model for non-invasive identification of perineural invasion in gallbladder carcinoma (GBC) based on preoperative enhanced CT imaging features.Methods:The clinical data of 180 patients who underwent curative-intent resection for gallbladder carcinoma at the First Affiliated Hospital of Xi′an Jiaotong University from January 2022 to December 2023 were retrospectively analyzed, including 61 males and 119 females with the age of (65.3±10.2) years old. The 180 patients were divided into a training set ( n=126) and a testing set ( n=54), and based on perineural invasion, the 126 patients in the training set were divided into the perineural invasion group ( n=33) and the non-perineural invasion group ( n=93), and the other 54 patients in the testing set, there were 15 patients with perineural invasion and 39 patients without perineural invasion. Clinical data such as gender, age, perineural invasion, carbohydrate antigen 19-9 (CA19-9) level and tumor stage were collected from patients. Multivariate logistic regression model was used to analyze the risk factors of perineural invasion in gallbladder carcinoma patients. The correlation between clinical variables and perineural invasion was ranked in order of importance using the "feature_importance" package in Python software. Then, we developed a random forest prediction model for perineural invasion in gallbladder carcinoma patients, and the area under the receiver operating characteristic (ROC) curve and confusion matrix were used to assess the predictive ability of the model. Results:Multivariate logistic regression model analysis showed that patients with CA19-9 >39.0 U/ml ( OR=5.165, 95% CI: 1.650-16.174), T3 stage ( OR=6.037, 95% CI: 1.571-23.197), T4 stage ( OR=9.996, 95% CI: 2.177-45.898), and lymph node metastasis ( OR=7.829, 95% CI: 2.705-22.627) were with a high risk of perineural invasion occurrence (all P<0.05). The top three variables in the order of the importance ranking were CA19-9, lymph node metastasis, and T stage. Combining the results of multivariate analysis and importance ranking, CA19-9, lymph node metastasis, and T stage were used to develop a random forest prediction model for perineural invasion in gallbladder carcinoma patients. The results of ROC curve analysis showed that the areas under curves of the random forest model in the training and testing sets were 0.8250 and 0.7667, respectively. The confusion matrix results showed that the sensitivity were 75.76% and 73.33%, the specificity were 80.65% and 76.92%, and the accuracy were 79.36% and 75.93% in the training and testing sets, respectively. Conclusion:Random forest prediction model based on preoperative enhanced CT image features can be used as a noninvasive means of identifying perineural invasion in patients with gallbladder carcinoma.
