Role of artificial intelligence applied to ultrasound in endometrial cancer: a systematic review

To synthesize the application of artificial intelligence (AI) in ultrasound imaging for the assessment of endometrial cancer, with a focus on methodological approaches and diagnostic accuracy in predicting different outcomes.

PubMed, Scopus, and Web of Science databases were searched from inception up to January 16, 2025. Studies applying AI to ultrasound imaging in the diagnosis, staging, and management of endometrial malignant pathology were included. Quality assessment of the retrieved studies was performed using the Quality Assessment Tool for Artificial Intelligence-Centered Diagnostic Test Accuracy Studies (QUADAS-AI). The protocol was registered in the PROSPERO database (registration record CRD42025648961).

Thirty studies were included: 18 (60%) distinguished between benign and malignant endometrial lesions, 4 (13.3%) focused on predicting lymph node metastases, 3 (10%) evaluated myometrial invasion, and 2 (6.6%) classified tumor risk. Additionally, 2 studies assessed disease-free survival (6.6%), while another developed a model for the automated identification of endometrial lesions (3.3%). According to QUADAS-AI, most studies were at high risk of bias for subject selection (eg, sample size not specified, imaging preprocessing not performed) (27/30, 90%) and the index test (no external validation) (27/30, 90%) domains, and at low risk of bias for the reference standard (target condition correctly classified by the reference standard) (29/30, 97%) and the workflow (reasonable time between index test and reference standard) (29/30, 97%) domains. Models were externally validated in 3/30 studies (10%), internally cross-validated in 3/30 (10%), internally hold-out validated in 13/30 (43.3%), and not validated in 11/30 (36.7%).

Published research on AI applications in ultrasound for endometrial cancer primarily focuses on developing classification models to distinguish benign from malignant endometrial lesions and to stage the disease. Overall, ultrasound-based AI models have demonstrated strong predictive performance. However, most studies are limited by small sample sizes and a lack of external validation.