The predictive model of risk of parastomal hernia based on CT radiomics and clinical indicators
LI Yun-bo
CHEN Lang
TAO Qian
LIU Chen
TANG Bo
FANG Shuang
XIAO Jing-jing
XIAO Wei-dong
QIU Yuan
Abstract：Objective To develop a predictive model for the occurrence of parastomal hemia(PSH)after colostomy and validate it.Methods A total of 131 rectal cancer patients who underwent permanent colostomy in our hospital from January 2016 to December 2020 were retrospectively analyzed and their preoperative clinical information and abdominal CT were collected.Patients were divided into PSH group(n=43)and nonPSH group(n=88)according to whether PSH occurred.Then,the PSH and nonPSH groups are divided into training and validation sets in the ratio of 7∶3,respectively.The preoperative abdominal CT images of the 3rd lumbar vertebral level were segmented to extract the radionics features,and at the same time,the preoperative clinical indicators were collected and screened.Three algorithms,support vector machine(SVM),decision tree(DT),and random forest(RF),were used to construct a prediction model by incorporating the screened clinical indicators and radionics features.The accuracy,sensitivity,specificity,and area under the ROC curve(AUC)of the models were calculated to evaluate the predictive efficacy of different models.Results In terms of clinical indicators,the total serum protein and BMI in the PSH group were significantly higher than those in the no-PSH group(P＜0.05).A total of 6 non-zero coefficient features were selected from the 107 extracted radionics features through LASSO regression.The AUCs of the prediction models built by SVM,DT,and RF were 0.820,0.854,and 0.790 respectively in the training set;they were 0.804,0.762,and 0.732 respectively in the validation set.Conclusion Using patients'preoperative clinical examination data and CT images can help identify high-risk groups for PSH and provide a reference for the prevention and clinical personalized diagnosis and treatment.
Keywords：parastomal hemiaradiomicsmachine learningprediction model
Publication Date：2024-10-01
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：5( 1174-1178 )