CT Multidimensional Radiomics Combined with Inflammatory Immune Score For Preoperative Prediction of Pathological Grade in Esophageal Squamous Cell Carcinoma.

Abstract

Rationale and objectives: Inflammation and immune biomarkers can promote angiogenesis and proliferation and metastasis of esophageal squamous cell carcinoma (ESCC). The degree of pathological grade reflects the tumor heterogeneity of ESCC. The purpose is to develop and validate a nomogram based on enhanced CT multidimensional radiomics combined with inflammatory immune score (IIS) for predicting poorly differentiated ESCC.

Materials and methods: A total of 266 ESCC patients from the retrospective study were included and randomly divided into a training set (N=186) and a validation set (N=80), and a complete data set (N=266), and overall survival was determined to follow up after surgery. The tumor imaging was segmented to form intratumoral and peritumoral 3 mm areas of 3D volume of interest (VOI) on CT arterial and venous phases, and 3404 radiomics features were extracted. Finally, the radiomics scores were calculated for arterial phase intratumoral (aInRads), peritumoral 3 mm (aPeriRads3), and venous phase intratumoral (vInRads), peritumoral 3 mm (vPeriRads3). Logistic regression was used to fuse the four cohorts of scores to form a Stacking. Additionally, sixteen inflammatory-immune biomarkers were analyzed, including aspartate aminotransferase to lymphocyte ratio (ALRI), aspartate aminotransferase to alanine aminotransferase ratio (AAR), neutrophil times gamma-glutamyl transpeptidase to lymphocyte ratio (NγLR), and albumin plus 5 times lymphocyte sum (PNI), etc. Finally, IIS was constructed using ALRI, AAR, NγLR and PNI. Model performance was evaluated by area under receiver operating characteristic curve (AUC), calibration curve, and decision curve analyse (DCA).

Results: Stacking and IIS were independent risk factors for predicting poorly differentiated ESCC (P<0.05). Ultimately, three models of the IIS, Stacking, and nomogram were developed. Compared with the Stacking and IIS models, nomogram achieved better diagnostic performance for predicting poorly differentiated ESCC in the training set (0.881vs 0.835 vs 0.750), validation set (0.808 vs 0.796 vs 0.595), and complete data set (0.857 vs 0.823 vs 0.703). The nomogram achieved an AUC of 0.881(95%CI 0.826-0.924) in the training set, and was well verified in the validation set (AUC: 0.808[95%CI 0.705-0.888]) and the complete data set (AUC: 0.857[95%CI 0.809-0.897]). Moreover, calibration curve and DCA showed that nomogram achieved good calibration and owned more clinical net benefits in the three cohorts. KaplanMeier survival curves indicated that nomogram achieved excellent stratification for ESCC grade status (P<0.0001).

Conclusion: The nomogram that integrates preoperative inflammatory-immune biomarkers, intratumoral and peritumoral CT radiomics achieves a high and stable diagnostic performance for predicting poorly differentiated ESCC, and may be promising for individualized surgical selection and management.

Availability of data and materials: The original manuscript contained in the research is included in the article. Further inquiries can be made directly to the corresponding author.