JMIR Publications

ABSTRACT

Background:

Several recent studies have investigated the diagnosis performance of radiomics-based machine learning (ML) models in identifying myometrial invasion (MI) in endometrial cancer (EC). However, a thorough synthesis of the available evidence regarding their accuracy remains lacking. Therefore, our study endeavors to systematically assess the diagnosis performance of radiomics-based ML approaches in identifying MI in EC, thereby providing evidence for noninvasive diagnosis tool development or improvement.

Objective:

Our study endeavors to systematically assess the diagnosis performance of radiomics-based ML approaches in identifying MI in EC, thereby providing evidence for noninvasive diagnosis tool development or improvement.

Methods:

PubMed, Cochrane Library, Embase, and Web of Science were thoroughly retrieved until November 26, 2024, for studies examining the utilization of radiomics-based ML in MI detection among the EC population. The risk of bias in eligible studies was evaluated through the Radiomics Quality Score (RQS). Our meta-analysis leveraged a bivariate random-effects model to evaluate diagnosis accuracy data presented in 2×2 contingency tables across validation sets.

Results:

19 studies comprising 4,373 EC patients were encompassed. The pooled estimates from the meta-analysis displayed the sensitivity and specificity of 0.79 (95% CI: 0.73-0.83) and 0.83 (95% CI: 0.79-0.86). Subgroup analysis revealed that deep learning (DL) models achieved the sensitivity and specificity of 0.81 (95% CI: 0.71-0.88) and 0.86 (95% CI: 0.76-0.92), whereas conventional machine learning (CML) models exhibited the sensitivity and specificity of 0.77 (95% CI: 0.69-0.83) and 0.81 (95% CI: 0.77-0.85).

Conclusions:

Radiomics-based ML models hold promise in predicting MI in EC, with DL approaches appearing to outperform CML methods. Nonetheless, the current body of evidence remains limited. Future research should incorporate more comprehensive imaging data and strive toward the development of intelligent and robust monitoring tools.