ENHANCED CT-BASED RADIOMICS FOR DISCRIMINATION PATHOLOGICAL SUBTYPES OF INVASIVE ADENOCARCINOMA

Background: This work aims at developing and validating enhanced CT-based Radiomics for preoperative discrimination of pathological subtypes of invasive adenocarcinoma. Methods: A retrospective analysis was carried out on 116 patients with invasive adenocarcinoma, being confirmed by surgical pathology. Preoperative chest enhanced CT and routine examination from January 2017 to December 2020 were performed. According to the postoperative pathology, they were divided into a low grade (micropapillary, solid) and medium-high grade (lepidic, acinar, nipple). We extracted imaging features by the AK software to establish the Radiomics signature (Rad-score). The prediction efficiency of low and moderate-high grades was evaluated by the ROC curve. A joint prediction model was used to combine Radiomics and independent clinical data. Nomogram and Calibration curves were used to assess the model efficiency. Results: The joint model had a good predictive effect for distinguishing pathological subtypes of invasive adenocarcinoma. In the training group, AUC was 0.96, sensitivity was 90.6%, specificity was 90.0%, and accuracy was 90.2%. In the testing group, AUC was 0.95, sensitivity was 100%, specificity was 90.5%, and accuracy was 94.1%. Nomogram based on Radscore, nature of tumor edge, had good performance in testing and training sets. Conclusion: The joint prediction model, combining the Radscore, CT value, and tumor margin established by chest CT of venous phase, was valuable for distinguishing low and moderate-high grade invasive adenocarcinoma before operation.