RadioPathomics Artificial Intelligence Model to Predict Tumor Regression Grading in Locally Advanced Rectal Cancer

Overview

In this study, investigators apply a radiopathomics artificial intelligence (AI) supportive model to predict neoadjuvant chemoradiotherapy (nCRT) response before the nCRT is delivered for the patients with locally advanced rectal cancer (LARC). The radiopathomics AI system predicts individual tumor regression grading (TRG) category based on each patient's radiopathomics features extracted from the Magnetic Resonance Imaging (MRI) and biopsy images. The predictive power to classify each patient into particular TRG category will be validated in this multicenter, prospective clinical study.

This is a multicenter, prospective, observational clinical study for validation of a radiopathomics integrated artificial intelligence (AI) system. Patients who have been pathologically diagnosed as rectal adenocarcinoma and defined as clinical II-III staging without distant metastasis will be enrolled from the Sixth Affiliated Hospital of Sun Yat-sen University, the Third Affiliated Hospital of Kunming Medical College and Sir Run Run Shaw Hospital Affiliated by Zhejiang University School of Medicine. All participants should follow a standard treatment protocol, including neoadjuvant concurrent chemoradiotherapy (nCRT), total mesorectum excision (TME) surgery and adjuvant chemotherapy. Images of Magnetic Resonance Imaging (MRI) and biopsy hematoxylin \& eosin (H\&E) stained slides of each patient should be available before nCRT treatment. The tumor region within these images would be delineated manually by experienced radiologists and pathologists. Further, the outlined images will be presented to the radiopathomics AI system to classify each participant into particular tumor regression grading (TRG) category. Here, the American Joint Committee on Cancer and College of American Pathologist (AJCC/CAP) 4-category TRG system is served as the standard. The actual TRG category of each participant will be confirmed based on pathologic assessment after TME surgery. Through comparisons of the predicted TRG and actual TRG category, investigators calculate the prediction accuracy, specificity and sensitivity as well as the F1 score. This study is aimed to develop a reliable and robust AI system to predict pathologic TRG prior to nCRT administration, facilitating response-guided precision therapy for patients with locally advanced rectal cancer.