Preoperative Assessment Of Histologic Grade In Colorectal Adenocarcinoma With CT-based Radiomics Approach

Background:Histologic grade of colorectal adenocarcinoma(CRAC)is a crucial factor for prognosis in patients with CRAC.Patients with high-grade CRAC showed higher risk in local recurrence after tumor resection,hence resulting in poorer prognosis compared to low-grade one.Accurately preoperative evaluation of histologic grade in CRAC would make for individualized treatment of patients and hence improving outcomes.At present,pathologic examination of colonoscopic biopsy samples is usually applied to preoperatively evaluate histologic grade of CRAC in clinical practice,but the accuracy of pathologic evaluation based on biopsy samples was not enough,due to superficiality,inadequacy,or poor orientation.Moreover,as an invasive examination,colonoscopic biopsy could lead to related side effects sometimes.Therefore,it is vital to explore and develop a non-invasively new approach to preoperatively evaluate histologic grade of CRAC,and the burgeoning radiomics,allowing high-throughput exploration of quantitative information from tumor images,may have such potential.Purpose:To explore the clinical value of applying CT-based radiomics approach for preoperative assessment of histologic grade in CRAC.Methods:This retrospective study was approved by our institutional review board,with the informed consent waived.This study enrolled 366 patients with CRAC from January 2008 to August 2015,which were divided into training dataset with 222 patients and validation dataset with 144 patients.Radiomics feature extraction was based on portal venous phase CT images of the enrolled patients,which included first-order statistics,size-and shape-based features,texture features and wavelet features.In training dataset,the least absolute shrinkage and selection operator(LASSO)method was applied to select radiomics features highly associated with histologic grade of CRAC and subsequently develop a radiomics signature.The correlation between radiomics signature and histologic grade was analyzed with Mann-Whitney U test in both training and validation dataset.The discrimination of radiomics signature in preoperatively evaluating histologic grade was investigated by the receiver operating characteristic(ROC)curve,and the area under the curve(AUC)with 95%confidence interval(Cl)was calculated.The calibration was analyzed with calibration curve.An independent validation dataset was designed to confirm the performance of the radiomics signature.We also performed a stratified analysis for further validating the performance of radiomics signature in rectal adenocarcinoma and colon adenocarcinoma.Results:After radiomics feature extraction and selection,radiomics signature was developed with 14 radiomics features.The radiomics signature demonstrated good performance to evaluate histologic grade of CRAC,with an AUC of 0.812(95%Cl:0.749-0.874)in training dataset and 0.735(95%Cl:0.644-0.826)in validation dataset.Stratified analysis fiurther confirmed the radiomics signature also showed discriminative ability for histologic grade in both rectal adenocarcinoma and colon adenocarcinoma,with AUC of 0.895(95%Cl:0.838-0.952)and 0.725(95%Cl:0.653-0.797)respectively.The radiomics signature demonstrated good calibration in each dataset as well.Conclusion:CT?based radiomics signature could be used as a non-invasively complementary tool for preoperative evaluation of histologic grade in CRAC,which may make contribution to personalized treatment.