Revolutionizing gastrointestinal cancer research with artificial intelligence: From precision patient stratification to real-world evidence

Table 1 Representative applications of artificial intelligence across the clinical research pipeline in gastrointestinal oncology

Research or products	Application scenarios		Tumor type	Main technologies/algorithms	Findings/strengths	Ref.
AI optimizes clinical trial design	Precise patient stratification	Multimodal integration	GIST	CE-CT WSI	Constructing a multimodal model to predict the RFS of GIST that outperforms unimodal models	Xiao et al[99], 2024
	Dynamic endpoint selection	AI predicts alternate endpoints	Pancreatic cancer	Kaplan-Meier, Cox regression and recursive partitioning analysis	By AI analysis, it was found that the combined assessment of CA19-9 response and MPR (TRG0-1) was a better predictor of pre-OS in patients with PDAC	[31]
AI improves patient recruitment & adherence	Intelligent matching system	EHR + NLP parsing	None	NLP	The development of NLP process enables automated extraction of tumor staging and diagnostic information across cancer types with high accuracy	Abedian et al[39], 2021
	Patient adherence management	AI remote follow-up personalized intervention	Gastric cancer	CT, DL	DL based on hybrid models may be a potential tool for predicting malnutrition in gastric cancer patients	Huang et al[46], 2024
Application of AI in efficacy and safety assessment	Automated imaging/pathology evaluation	Image recognition	Colorectal cancer liver metastasis	Fully convolutional networks	The method can effectively improve the work efficiency, but it needs further improvement in segmentation accuracy and consistency	Vorontsov et al[53], 2019
		Digital pathology	Colon	WSI	This method can accurately predict molecular features such as gene mutations, copy number changes, MSI, and protein expression in colon cancer patients	Ding et al[60], 2022
	Adverse event forecasting	AI risk model	None	DySPred	The model demonstrated good stability, was able to predict toxicity trends in different populations and cancer types with small samples, and could be effectively compared with transcriptional changes of small molecule antitumor drugs	Yan et al[62], 2025
		Value and limitations	Colorectal cancer	DL	Models perform well on specific datasets, but they do not generalize well enough to perform consistently well across many different clinical scenarios	Höhn et al[63], 2023
RWE and AI	RWD into RWE	Fusion of multiple data sources, AI improves data cleaning efficiency	None	MSK-CHORD	Demonstrates the feasibility of automated annotation from unstructured notes and its utility in predicting patient prognosis	Jee et al[74], 2024
	AI bridging clinical trials & practices	Model generalizability validation	None	AI	AI has outstanding advantages in the processing of multiple sets of data and decision making for precise tumors, but still falls short in the areas of limited data on rare cancers, clinical validation needs, and regulatory issues	Vyas et al[78], 2025

AI: Artificial intelligence; GIST: Gastrointestinal stromal tumors; CE-CT: Contrast-enhanced cranial computed tomography; WSI: Whole-slide image; RFS: Recurrence-free survival; MPR: Major pathologic response; OS: Overall survival; PDAC: Pancreatic ductal adenocarcinoma; EHR: Electronic health record; NLP: Natural language processing; CT: Computed tomography; DL: Deep learning; MSI: Microsatellite instability; RWD: Real-world data; RWE: Real-world evidence.