A systematic review of radiomics for predicting treatment response and survival in locally advanced cervical cancer: positive results or optimistic illusions?

ObjectiveTo evaluate the value and study quality of radiomics studies in locally advanced cervical cancer (LACC) for treatment response and survival prediction.MethodsA systematic review was conducted, according to PRISMA guidelines. The protocol was prospectively registered (PROSPERO: CRD42023397117). Four databases of PubMed, Embase, Web of Science, and Scopus were searched for articles on treatment response and survival prediction in LACC. Methodological quality was assessed using the Radiomics Quality Score (RQS), and the risk of bias for prediction models and predictors was assessed using the Prediction model Risk of Bias Assessment Tool (PROBAST) and Quality in Prognosis Studies (QUIPS) tools, respectively, independently evaluated by two reviewers.ResultsA total of 30 studies were included. The median RQS was 10 (range - 6 to 17), with 13 studies scoring above 30%. In terms of PROBAST, all included studies were assessed as having a high risk of bias (ROB) overall. Regarding QUIPS, only two studies were assessed as having a low risk of bias. Twenty-three and nine studies reported results for predicting survival and treatment response, respectively, and most of these radiomics models or features had good predictive performance, but validation is lacking and none of the studies provide discussion of biological correlation. Only a few of the studies using multiple scanners reported an effect of harmonization strategy on results, and two studies using the ComBat harmonization method reported inconsistent results.ConclusionsRadiomics for predicting treatment response and survival in LACC is still in its infancy, and the available results must be treated cautiously. Future studies should focus more on standardized radiomics workflows, validation of predictive models, and reporting the correlation of biological information with predicted outcomes.