# Radiomics for Detection and Differentiation of Intrahepatic Cholangiocarcinoma: A Systematic Review and Meta-Analysis

## TL;DR

This study reviews how radiomics and AI can help detect and distinguish intrahepatic cholangiocarcinoma from other liver tumors, showing promising diagnostic accuracy.

## Contribution

The study provides a systematic review and meta-analysis of radiomics-based AI models for detecting intrahepatic cholangiocarcinoma.

## Key findings

- Radiomics-based AI models correctly identified intrahepatic cholangiocarcinoma in 77% of cases.
- These models ruled out the cancer in 88% of cases, with CT-based models performing best.
- The study highlights the need for standardized protocols and external validation before clinical adoption.

## Abstract

Intrahepatic cholangiocarcinoma is a rare but highly aggressive cancer that arises from the bile ducts within the liver and is frequently diagnosed at an advanced stage, limiting treatment options and survival. This study aimed to evaluate whether advanced computer-based analysis of medical images called radiomics, a method that extracts hidden patterns from scans, can improve the detection and differentiation of this cancer from other liver tumors. We analyzed results from 20 studies, including 8746 patients in which artificial intelligence models examined CT, MRI, or ultrasound images. Overall, these models correctly identified the cancer in about 77% of cases and correctly ruled it out in about 88% of cases, with CT-based models showing the strongest performance. These findings suggest that computer-assisted image analysis may support earlier and more accurate diagnosis, helping clinicians make better treatment decisions and potentially improving outcomes for patients with liver tumors.

Background: Intrahepatic cholangiocarcinoma (ICC) is an aggressive primary liver malignancy with limited survival, largely due to delayed diagnosis, recurrence and limited effective therapeutic options. Radiomics- and artificial intelligence (AI)-based imaging models have emerged as promising tools to improve noninvasive detection and differentiation of ICC. We conducted a systematic review and meta-analysis to evaluate the diagnostic performance of radiomics-based AI models for ICC. Methods: A systematic search of PubMed, Embase, Scopus, and the Cochrane Library was performed from inception through 2025 in accordance with PRISMA guidelines. Studies assessing radiomics- or AI-based models derived from CT, MRI, PET, or ultrasound for differentiation of ICC from other hepatic lesions were included. Pooled sensitivity, specificity, positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were estimated using a bivariate random-effects model. Study quality and risk of bias were assessed using the Radiomics Quality Score (RQS) and QUADAS-2 tools. Results: Twenty retrospective studies comprising 8746 participants were included. Across pooled validation and test datasets, radiomics-based AI models demonstrated a pooled sensitivity of 0.77 (95% CI, 0.69–0.84) and specificity of 0.88 (95% CI, 0.78–0.94) for differentiating ICC from non-ICC hepatic lesions. The pooled PLR was 6.81 (95% CI, 3.51–13.2), and the pooled NLR was 0.23 (95% CI, 0.09–0.61). CT-based models showed higher diagnostic performance compared with MRI and ultrasound. Subgroup and meta-regression analyses identified imaging modality, contrast phase, segmentation strategy, and validation approach as contributors to interstudy heterogeneity. The overall methodological quality demonstrated a mean Radiomics Quality Score (RQS) of 14.0 (range 11–24), corresponding to approximately 39% of the maximum achievable score. External validation cohorts were incorporated in 60% of the studies, although adherence to standardized feature reproducibility frameworks varied. Conclusions: Radiomics-based AI models demonstrate clinically meaningful diagnostic accuracy for noninvasive differentiation of ICC and may complement conventional imaging in preoperative evaluation. Prospective, multicenter studies with standardized imaging protocols and rigorous external validation are required before routine clinical adoption.