Radiomics and machine learning models for diagnosing microvascular invasion in cholangiocarcinoma: a systematic review and meta-analysis of diagnostic test accuracy studies

Purpose

We aimed to systematically assess the value of radiomics/machine learning (ML) models for diagnosing microvascular invasion (MVI) in patients with cholangiocarcinoma (CCA) using various radiologic modalities.

Methods

A systematic search of was conducted on Web of Sciences, PubMed, Scopus, and Embase. All the studies that assessed the value of radiomics models or ML models along with the use of imaging features were included. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) criteria and METhodological RadiomICs Score (METRICS) were used for quality assessment. Pooled estimates for the diagnostic performance of radiomics/ML models were calculated. I-squared was used to assess heterogeneity, and sensitivity and subgroup analyses were performed to find the sources of heterogeneity. Deeks' funnel plots were used to assess publication bias.

Results

11 studies were included in the systematic review with only one study being about extrahepatic CCA. According to the METRICS, the mean score was 62.99 %. Meta-analyses were performed on intrahepatic CCA studies. The meta-analysis of the best ML models revealed an AUC of 0.93 in the training cohort and an AUC of 0.85 in the validation cohort. Regarding the best radiomics model, the AUC was 0.85 in the training cohort and 0.81 in the validation cohort.

Conclusion

Radiomics/ML models showed very good diagnostic performance regarding MVI diagnosis in patients with intrahepatic CCA and may provide a non-invasive method for this purpose. However, given the high heterogeneity and low number of the included studies, further multi-center studies with prospective design and robust external validation are essential.