Abstract / Summary
Histopathological growth patterns (HGPs) in colorectal liver metastases (CRLM) significantly influence patient prognosis and response to systemic therapy. While the desmoplastic HGP is associated with better outcomes, the replacement HGP often indicates a more aggressive phenotype. Current preoperative imaging cannot distinguish these patterns, and biopsies are often unreliable due to limited sampling of the tumor–liver interface. This systematic review and meta-analysis evaluates the diagnostic performance of radiomics/deep learning models for the non-invasive prediction of HGPs based on imaging data. This study was conducted following PRISMA guidelines, searching 4 databases up to August 30, 2026. Studies developing radiomics or deep learning models to predict HGPs in CRLM patients using preoperative CT or MRI were included. Methodological quality and risk of bias were assessed using the RQS 2 and QUADAS-2 tools. Pooled sensitivity and specificity were calculated using a random-effects model. Eight retrospective reports representing seven independent study populations were included in the systematic review; seven independent populations comprising 724 unique patients were included in the quantitative synthesis. The pooled sensitivity and specificity were 0.86 (95% CI: 0.81–0.90) and 0.77 (95% CI: 0.64–0.87), respectively. The SROC curve yielded an AUC of 0.845. Subgroup analysis did not reach statistical significance in most analyses, and its findings should be interpreted with caution given the limited number of studies available. The mean RQS 2.0 score was 17.1 (31%), indicating that while technical development is standardized, prospective validation and clinical application measures are currently lacking. Imaging-based AI models showed encouraging discriminatory performance for non-invasive HGP classification. However, the available evidence does not demonstrate clinical utility or improvement in patient management or outcomes. Prospective multicenter validation and decision-impact studies are required before clinical implementation can be considered.