Radiomics in prognostication and tumor grading for gastrointestinal cancer: a scoping review

Abstract Background GI cancers are a major cause of cancer-related death in the world with an incident of 4.8 million and a mortality of 3.4 million patients every year. Extracting quantitative imaging features through radiomics can be better at prognostication than traditional clinical tools. Methods This study employed a two-stage approach to map and synthesize the current evidence in the field of radiomics. We performed a scoping review and bibliometric analysis according to PRISMA-ScR. Among 1620 records in Scopus, PubMed, and Web of Science, 42 studies were included in accordance to the inclusion criteria. Biblioshiny examined the trends of publications and global collaborations. Results The 2020–2025 (peak: 244 annually) publications lead to a good collaboration between China and the USA and Europe. All the researches showed that radiomics outperforms the clinical-only models (AUC improvement: 0.05–0.18). Studies have concentrated on colorectal (31.0%), gastric (28.6%), and esophageal (19.0%) cancers; pancreatic/hepatobiliary cancers are underrepresented (9.5 each). Methodological heterogeneity was very high: Extractions were made of 10–1000 features, standardized protocol was only used by 35.7 percent, and external validation was done by 19.0 percent. The use of deep learning grew by 25 (2020–2022) to 58 (2023–2025). Clinical integrations came out using radiomics-clinical nomograms (54.8%) and web-based tools (27.0%). Conclusions Radiomics is always seen to improve the prognostication of GI cancer but it needs methodological standardization, external large-scale validation, and trials before it is clinicalized. Pancreatic and hepatobiliary cancers require a priority research.