Research on Intelligent Screening and Decision-making for Neoadjuvant Therapy in Locally Advanced Gastric Cancer Based on Multi-omics Integration

Overview

In this study, investigators utilize a radiopathomics integrated Artificial Intelligence (AI) supportive system to predict tumor response to neoadjuvant chemoradiotherapy (nCRT) before its administration for patients with locally advanced gastric cancer (LAGC). By the system, the postoperative tumor regression grade (TRG) of the participants will be identified based on the radiopathomics features extracted from the pre-nCRT Enhanced CT and biopsy images. The ability to predict TRG will be validated in this multicenter, prospective clinical study.

This is a multicenter, prospective, observational clinical study for validation of a radiopathomics artificial intelligence (AI) system. Patients who have been diagnosed with gastric adenocarcinoma by pathology and defined as clinical stage II-IVa without distant metastasis by enhanced CT scan will be enrolled from the Second Affiliated Hospital of Zhejiang University, the First Affiliated Hospital of Zhejiang University, Shangyu People's Hospital of Shaoxing City and Zhejiang Cancer Institute \& Hospital. All participants should adhere to a highly standardized treatment protocol, which involves receiving either 2-4 courses of standard neoadjuvant chemotherapy based on 5-FU + platinum, or 2-4 courses of neoadjuvant chemotherapy based on 5-FU + platinum combined with trastuzumab, or 2-4 courses of neoadjuvant chemotherapy based on 5-FU + platinum combined with anti-PD-L1 therapy. Following the neoadjuvant treatment protocol, participants will undergo a D2 radical gastrectomy for gastric cancer. The enhanced CT and biopsy examination should be completed before the nCRT and the images will be subjected to the manual delineation of the tumor regions of interest (ROI) by experienced radiologists and pathologists. Subsequently, the enhanced CT and biopsy images outlined will be used in the radiological pathology AI system to generate predicted responses (predicted postoperative TRG grading) for individual patients, while actual responses (confirmed postoperative TRG grading) will be diagnosed in surgical resection specimens. Through comparisons of the predicted responses and true pathologic responses, investigators calculate the prediction accuracy, specificity, sensitivity as well as the Area Under Curve (AUC) of Receiver Operating Characteristic (ROC) curves. The aim of this study is to verify the high accuracy and robustness of the radiological pathology AI system in predicting postoperative TRG grading in individuals before nCRT, which will promote further precise treatment of locally advanced cancer patients.