Prediction Of Microsatellite Instability In Colorectal Cancer Based On PET/CT Radiomics:a Multicenter Cohort Study

BackgroundColorectal Cancer(CRC)is the third most prevalent malignant tumor globally and the second primary cause of cancer-related deaths.In China,CRC ranks second in incidence and fifth in mortality.The prognosis of CRC is significantly influenced by tumor histological type,cell differentiation,RAS/BRAF gene mutation,and microsatellite instability.Microsatellite instability has been identified as a significant molecular pathogenic mechanism of CRC,which holds immense importance in the classification,treatment,and prognosis of CRC.The detection methods of microsatellite instability are based on biopsy specimens.Due to different detection methods,the results are different,which are interfered by factors such as sample,staining intensity,tumor cell heterogeneity,and special pathological morphology.Radiomics,as an emerging feature extraction quantitative analysis technology,has the characteristics of high throughput,high specificity,non-invasiveness,and economy.It uses the patient’s preoperative imaging examination data to extract imaging feature data from medical images to obtain high-dimensional image data of tumor phenotype and microenvironment.Statistical or artificial intelligence techniques were used to analyze and interpret these data to construct predictive models to predict microsatellite instability in CRC non-invasive and assist in treatment decisions.ObjectiveBased on the features of preoperative abdominal enhanced CT and 18F-FDG PET/CT images and clinical information,a prediction model of microsatellite instability in CRC patients was constructed to predict non-invasive microsatellite instability of CRC and assist clinical decision-making.Materials and MethodsThe clinical information and imaging data of colorectal cancer in Guangdong Provincial People’s Hospital,Zhuhai People’s Hospital,and the First Affiliated Hospital of Wenzhou Medical University from March 2012 to June 2021 were retrospectively collected.The data from the first hospital were used as the training and test cohorts(7:3),and the abdominal contrast-enhanced CT and PET/CT radiomics features were extracted by manual segmentation method,respectively.Consistency assessment,univariate analysis,least absolute shrinkage and selection operator(LASSO),and multivariate logistic regression analysis were used for feature selection.Nomogram was used to visualize the model.The data from two other hospitals were used as the external validation cohort to verify the model’s external applicability and generalization ability.ResultsA total of 10 abdominal contrast-enhanced CT features and 9 PET/CT features were screened.The area under the curve(AUC)of the training cohort based on abdominal-enhanced CT features,PET/CT features,and combined features were 0.717,0.797,and 0.801,respectively.The test cohort was 0.642,0.775,and 0.744,respectively.The PET/CT feature model in the test cohort had the highest AUC,and the AUC,sensitivity,specificity,and accuracy of the combined model were 0.874,0.909,0.824,and 0.832,respectively.Decision curve analysis,NRI,and IDI all show that the joint model is optimal.Using the external validation cohort,the AUC,sensitivity,specificity,and accuracy of the combined model were the highest,which were 0.747,0.545,0.868,and 0.848,respectively.ConclusionsThe visual Nomogram based on PET/CT radiomics features combined with clinical information has a good performance in predicting microsatellite instability in CRC patients,which can assist in the diagnosis and treatment planning of CRC patients.