Answers are grounded in this paper and cite their sources.
Aim: Neoadjuvant therapy for locally advanced rectal cancer (LARC) yields variable patient responses. Artificial intelligence (AI) and digital pathology offer a novel approach to preoperatively predict treatment outcomes using histomorphological features, potentially guiding organ-preserving strategies. This scoping review maps the current literature on AI-driven neoadjuvant response prediction in LARC to evaluate algorithmic performance and identify methodological gaps. Methods: Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for scoping reviews (PRISMA-ScR) guidelines, a structured search of PubMed, Embase, and the Cochrane Library was conducted for studies published within the last 10 years. Studies applying AI algorithms to pretreatment whole-slide images (WSIs) to predict neoadjuvant response in LARC were included. Results: Eleven publications were included. Ten studies used retrospective cohorts only, whereas one study also included a prospective validation cohort of 100 patients. Algorithmic frameworks predominantly featured classical machine learning, using software extraction of predefined features, and deep learning for autonomous pixel analysis. Across heterogeneous endpoints and cohorts, standalone classical machine-learning models reported external validation area under the receiver operating characteristic curve (AUC) of 0.811-0.908 (prospective: 0.733), a hybrid microenvironment model reported external AUCs of 0.643-0.656 (prospective: 0.630), and standalone deep-learning models reported an external AUC of 0.723 (internal: 0.710-0.81). Critical limitations across the field include heterogeneous clinical target endpoints, inconsistent baseline staging, and a reliance on “black box” algorithms lacking true biological explainability. Conclusion: AI-driven digital pathology demonstrates strong potential for predicting neoadjuvant treatment response in LARC. However, safe clinical implementation will require standardized endpoints, multimodal integration, and prospective multi-centre validation of algorithms anchored in verifiable tumour biology.
No takes yet. Share an insight, caveat, or question.
Allen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: