AI Ultrasound Models Show High Accuracy in Ovarian Cancer Diagnosis

Artificial intelligence (AI) models using ultrasound imaging demonstrated strong diagnostic performance for ovarian cancer in both internal and external validation datasets, with the potential to outperform sonographers, according to a study published in Frontiers in Oncology.¹

Can AI Improve Early Ovarian Cancer Detection?

Subtle or non-specific ovarian cancer symptoms often result in delayed clinical presentation. Because of this, the researchers underscored the need for earlier, more accurate diagnostic approaches to reduce mortality, improve treatment outcomes, and minimize unnecessary surgical procedures for patients presenting with ovarian masses. Conventional diagnostic modalities include imaging tools such as CT or MRI, serum biomarkers, pathological biopsy, and ultrasound.

“While these diagnostic methods offer valuable insights, their limitations underscore the urgent need for innovative solutions to improve diagnostic precision and reliability,” the authors wrote.

More recently, AI has shown promise in enhancing diagnostic performance, particularly in the analysis of medical imaging such as ultrasound. Using machine learning and deep learning technologies, AI can extract complex patterns from imaging data and provide quantitative assessments of radiographic features often invisible to the human eye. A prior study reported that AI-assisted ovarian cancer diagnostics can achieve sensitivity and specificity rates as high as 81% and 92%, respectively, outperforming traditional image-based approaches.²

However, AI models also face notable challenges, highlighting the need for systematic evaluation to clarify performance variability across studies.¹ To address this gap, the researchers conducted a meta-analysis to assess the diagnostic performance of AI-based ultrasound models for the initial diagnosis of ovarian cancer and to compare their performance with that of sonographers.

They searched PubMed, Web of Science, Embase, and the Cochrane Library for eligible studies through February 2025. Using pathology as the reference standard, the included studies employed AI algorithms to analyze ultrasound images from patients with suspected ovarian cancer.

The researchers used bivariate random-effects models to aggregate sensitivity, specificity, and area under the curve (AUC) from internal validation sets, external validation sets, and sonographers. Sensitivity reflected the probability of correctly identifying malignant cases, specificity represented the likelihood of accurately identifying non-malignant cases, and AUC served as a composite metric for diagnostic discrimination. Additionally, the methodological quality was assessed using a modified version of the Quality Assessment of Diagnostic Accuracy Studies-2 tool.

How Accurate Is AI-Assisted Ultrasound for Diagnosing Ovarian Cancer?

The initial search identified 302 eligible studies, of which 18 met all criteria for final inclusion. This comprised 17 studies in the internal validation set involving 22,697 total patients, images, or lesions (range, 1-7995) and 3 studies in the external validation set involving 2297 patients (range, 2-662). Published between 1999 and 2024, 13 studies were retrospective, and 5 were prospective.

Radiomic and clinical AI models were used in 8 studies, while radiomic-only models were used in 10. Similarly, deep learning methods were employed in 8 studies, with the remaining 10 using traditional machine learning approaches.

In internal validation sets, AI demonstrated a sensitivity (95% CI, 0.88-0.98) and specificity (95% CI, 0.89-0.98) of 0.95, yielding an AUC of 0.98. Meanwhile, 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. By comparison, sonographers exhibited a sensitivity of 0.83 (95% CI, 0.62-0.94), a specificity of 0.84 (95% CI, 0.79-0.88), and an AUC of 0.87.

The researchers emphasized that these findings suggest ultrasound-based AI models significantly outperform sonographer diagnostics. They also observed high heterogeneity across studies and, through a meta-regression analysis, determined that this was primarily attributed to the differences between image-based and patient-based analysis methods (P = .01).

What Further Evidence Is Needed to Support the Clinical Use of AI-Based Ultrasound?

The researchers acknowledged several limitations, including that most analyzed studies were retrospective, which may introduce bias. Additionally, when multiple AI algorithms were reported within a study, only the optimal model was selected for analysis, potentially leading to an overestimation of diagnostic performance. However, they expressed confidence in their findings and highlighted the need for further research.

“…these results indicate the potential for AI integration into clinical practice,” the authors wrote. “Further studies with external, multicenter, prospective head-to-head design are still needed.”

References

1. Li R, Lei J, Tang X, et al. Artificial intelligence based on ultrasound for initial diagnosis of malignant ovarian cancer: a systematic review and meta-analysis. Front Oncol. 2025;15:1626286. doi:10.3389/fonc.2025.1626286
2. Xu HL, Gong TT, Liu FH, et al. Artificial intelligence performance in image-based ovarian cancer identification: a systematic review and meta-analysis. EClinicalMedicine. 2022;53:101662. doi:10.1016/j.eclinm.2022.101662