Postoperative Survival Prediction In Colorectal Cancer

BackgroundColorectal cancer(CRC)is a malignant tumor with the highest incidence rate and mortality.TNM staging system is still difficult to provide sufficient postoperative prognosis information for CRC patients.More accurate prognostic methods are still warranted to assist lin individualized treatment strategy.The emerging results of radiomics with prognostic utility have been found in a variety of solid malignancies,but it is still rare in the research and application in the context of prognosis of CRC.Recent results have shown that the heterogeneity of CRC can reflect the prognosis of patients,but the classic radiomics method still has limitations in the quantification of spatial heterogeneity based on the extraction of average values of radiomic features based on the whole region of the tumor.In the field of digital pathology,ecosystem diversity analysis has been used to quantify the spatial heterogeneity of tumors and assist in the prognosis prediction of malignant tumors.Therefore,it is urgent to investigate whether integrating the analysis method of tumor ecosystem diversity with radiomics could enable the comprehensively quantification of the spatial heterogeneity of CRC and reflect the postoperative prognosis of CRC patients.Materials and MethodsWe retrospectively identified 409 consecutive patients with stage Ⅰ-Ⅲ CRC who had underwent radical resection in our institute,followed by the randomly assignment of theses patients into the training cohort and internal validation cohort.A total of 103 patients from an external institute were included as the external validation cohort.Both classical radiomics features and EcoRad features were extracted from the CT images of the CRC patients,with the latter extracted leveraging the analysis of the diversification of tumor ecosystem.A classical radiomics signature and an EcoRad signature were constructed respectively using the classical radiomics features and EcoRad features.Multivariate Cox regression was used to build a prediction model integrating the classical radiomics signature,EcoRad signature,T stage and N stage,with its prediction performance demonstrated with respect to the discrimination(accuracy and C-index),calibration(calibration curve)and clinical usefulness(Decision curve analysis,DC A).ResultsThe classic radiomics signature and the EcoRad signature are both prognostic factors independent of clinicopathological factors.For OS prediction,both signatures showed good prediction performance.In the meanwhile,the prediction performance of the EcoRad signature was better than that of the classical radiomic signature,with the C-index of the Ecorad signature being 0.765(95%CI:0.745-0.785)and the C-index of the classical radiomic signature being 0.679(95%CI:0.659-0.699)in external validation set.The constructed prediction model integrates the classical radiomic signature,the EcoRad signature,the T and N stage.Good discrimination performance was shown by the prediction model,with the C-index being 0.746(95%CI:0.7220.770)among external validation patients.Good calibration and clinical utility was also demonstrated by the calibration curve and DCA curve of the prediction model.ConclusionThis study establishes a method to fully quantify the spatial heterogeneity of CRC through coupling tumor ecosystem analysis with radiomics approach based on CT images,suggesting this approach could effectively predict OS and serve as a supplementary tool to the staging system for survival prediction among CRC patients.