Inteligencia artificial basada en ultrasonido para el diagnóstico inicial de cáncer de ovario maligno: una revisión...

This meta-analysis aimed to evaluate the diagnostic performance of artificial intelligence (AI) in ultrasound imaging for the initial diagnosis of malignant ovarian cancer, comparing its performance to that of sonographers.

A systematic literature search was conducted in PubMed, Web of Science, Embase, and the Cochrane Library up to February 2025. Inclusion criteria targeted studies employing AI algorithms to analyze ultrasound images in patients with suspected ovarian cancer, using pathology as the reference standard. Bivariate random-effects models were utilized to aggregate sensitivity, specificity, and area under the curve (AUC). The methodological quality of the included studies was assessed using a modified version of the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool.

Eighteen studies encompassing a total of 22,697 total patients/images/lesions were analyzed. AI demonstrated a sensitivity of 0.95 (95% CI: 0.88-0.98) and specificity of 0.95 (95% CI: 0.89-0.98) in internal validation sets, yielding an AUC of 0.98. In external validation, sensitivity was 0.78 (95% CI: 0.56-0.91) and specificity was 0.88 (95% CI: 0.76-0.95), with an AUC of 0.91. In comparison, sonographers exhibited a sensitivity of 0.83 (95% CI: 0.62-0.94), specificity of 0.84 (95% CI: 0.79-0.88), and an AUC of 0.87. These results indicate that ultrasound-based AI significantly outperforms sonographer diagnostics. Meta-regression analysis indicated that the heterogeneity was primarily attributed to the analysis method (image-based vs. patient-based, specificity P = 0.01).

AI based on ultrasound diagnosis demonstrates excellent performance for malignant ovarian cancer detection, with potentially superior performance compared to sonographers. Despite high heterogeneity across studies and the observed publication bias, these results indicate the potential for AI integration into clinical practice. Further studies with external, multicenter prospective head-to-head design are still needed.