Deep Learning for Prediction of Neoadjuvant Chemoradiotherapy Response in Locally Advanced Rectal Cancer: A Retrospective Multi-Cohort Pathomics Study

Background: Predicting the neoadjuvant chemoradiotherapy (nCRT) response for patient with locally advanced rectal cancer (LARC) would facilitate therapeutic decisions. Yet, the approach to predict the treatment response before nCRT remains challenging. Here, we develop a multicenter deep learning model based on the pretreatment whole slide image (WSI) of hematoxylin and eosin (H&E) stained biopsy to distinguish the pathological complete response (pCR) from the LARC patients. Methods: Consecutive patients with WSI of endoscopic biopsy H&E-stained slides were included from two hospitals as a training cohort (n = 303) and a validation cohort (n = 154), respectively. Tissue regions of each WSI of H&E-stained biopsy slides were automatically segmented and subjected to extract deep features to construct a deep signature by a VGG-16 convolutional neural network model. A prediction model combining the deep signature with clinicopathologic predictors was developed and validated to identify the pCR group. Receiver operating characteristic (ROC) curves and decision curve analysis (DCA) was used to calculate the predictive performance. Findings: The combined model showed good discrimination of the pCR group in two cohorts [the average area under ROC (95% confidence interval): 0.87 (0.83-0.91) in the training cohort, and 0.79 (0.71-0.86) in the independent validation cohort]. Besides, DCA proved that the combined model was clinically valuable. Interpretation: The deep learning model based on biopsy WSI indicates that it might offer a simple preoperative tool to predicts nCRT treatment response for LARC patients. Funding Information: This study was funded by research grants from the National Natural Science Foundation of China [82001986], National Science Fund for Distinguished Young Scholars [81925023], National Key Research and Development Program of China [2017YFC1309102], the Applied Basic Research Projects of Yunnan Province, China [2019FE001-083 and 2018FE001-065], Yunnan digitalization, development and application of biotic resource [202002AA100007]. Declaration of Interests: The authors declare that there are no conflicts of interest. Ethics Approval Statement: The ethics committees of the Sixth Affiliated Hospital of Sun Yat-sen University (SYSU6) and Yunnan Cancer Hospital (YNCH) both approved the multicenter retrospective study. The board waived the requirement for informed consent because of the study’s retrospective nature. All data in the study were de-identified and anonymized.