Artificial intelligence-based model for recurrence prediction after local excision on whole slide images in T1 low rectal cancer.

Abstract

24 Background: According to current guidelines, patients with resected specimens showing high-risk features are recommended for additional surgery after local excision of T1 colorectal cancer, despite the low incidence (~7%) of recurrence. However, surgical resection in patients with low rectal cancer (RC) is challenging and may compromise anal function, leading to a low quality of life. To reduce unnecessary surgical resection for these patients, we utilized artificial intelligence to develop and validate a prediction model for the risk of recurrence in T1 low rectal cancer patients. Methods: H&E-stained whole slide images (WSIs) were scanned for local excision (endoscopically or transanally minimally) specimens of 507 consecutive patients with T1 low rectal cancers that were locally resected at 4 hospitals between 2005 and 2015. The area under the receiver operating characteristic curve (AUROC), specificity and sensitivity were used to evaluate the performance of the model for the risk of recurrence, and an external validation cohort was to verify the applicability of the model. Results: We constructed a prediction model using convolutional neural networks (CNN) without incorporating clinical features. The model yielded good discrimination and calibration, achieving a 5-year recurrence-free survival AUROC of 0.90 (95% confidence interval [CI]: 0.86–0.93), sensitivity of 0.91 (95% CI: 0.84–0.96), and specificity of 0.82 (95% CI: 0.78–0.88) through fivefold cross-validation. Additionally, the AI avoided 25.7% of unnecessary additional surgery compared to the current guidelines. Conclusions: We proposed a novel prediction model for the risk of recurrence in T1 low RC patients to assist physicians in determining whether additional surgery is required after local excision of T1 low RC.