# Radiomics and artificial intelligence-based prediction of tumor response in digestive system neoplasm: a systematic review and meta-analysis

## TL;DR

This paper reviews how radiomics and AI can predict tumor outcomes in digestive cancers, showing strong effectiveness across various diagnostic and prognostic applications.

## Contribution

The study provides a systematic review and meta-analysis of AI and radiomics applications in digestive system tumors, highlighting their diagnostic and prognostic potential.

## Key findings

- AI significantly improves detection in upper GI endoscopy, colonoscopy, and capsule endoscopy.
- Radiomic models effectively predict tumor response and recurrence in colorectal and gastric cancers.
- AI-based strategies accurately assess patient risk and predict molecular tumor types and survival.

## Abstract

Radiomics and artificial intelligence (AI) are progressively gaining recognition for predicting tumor response, recurrence, and prognosis in gastrointestinal tumors. The current review singled out the diagnostic and prognostic potential of AI and radiomics in the whole GI tract.

Out of 120 ongoing studies from the year 2016 to 2025, the following applications were covered: endoscopy, colonoscopy, capsule endoscopy, intraoperative guidance, CT/MRI radiomics, and molecular/histopathology AI models. The performance across studies was assessed by meta-analysis using random-effects modeling that incorporated inverse variance methods. Results from the analysis of heterogeneity (I2), publication bias (funnel plots, Egger's test), methodological quality (Radiomics Quality Score, RQS), and risk of bias (PROBAST) were reported.

The use of AI in detection and diagnosis assisted with the endoscopy of the upper gastrointestinal tract (OR = 16.12, 95% CI: 7.72–33.65), colonoscopies for colorectal polyps (OR = 12.0, 95% CI: 10.26–14.03), and capsule endoscopy (OR = 10.16, 95% CI: 8.32–12.4) and was proven to be very effective. Intraoperative guidance also was proven to be an effective surgical decision-making tool (OR = 8.12, 95% CI: 7.12–9.26), whereas an AI-based strategy for patient risk assessment predicted the occurrence of lymph node metastasis, molecular tumor types, and patient survival (OR = 9.62, 95% CI: 7.93–11.66). Radiomic models forecasted tumor responses and relapses in rectal/colorectal (OR = 10.48, 95% CI: 9.66–11.36), gastric/esophagogastric/esophageal cancers (OR = 10.81, 95% CI: 9.89–11.82), molecular/histopathology datasets (OR = 11.62, 95% CI: 10.42–12.95), and CT/MRI recurrence/prognosis models (OR = 10.59, 95% CI: 9.52–11.79). The RQS assessment indicated moderate-to-high methodological quality, and the PROBAST evaluation revealed a low-to-moderate risk of bias.

Validation through prospective multicenter studies and reporting that has been standardized is the key to clinical reliability enhancement and backed-up precision oncology implementation.