Ovarian Cancer: Diagnostic Challenges and Current Practices

Ovarian cancer is the most lethal gynecologic malignancy, with a five-year survival rate below 50%, primarily due to frequent late-stage diagnoses.¹ Benign adnexal masses, common across all age groups, are often asymptomatic and typically detected incidentally during imaging for other purposes.² Nearly 10% of asymptomatic postmenopausal women have an ovarian lesion, though only 1% are malignant.³ Managing these patients presents significant diagnostic challenges, requiring accurate triage to balance the risks of missing malignancies against unnecessary surgeries.

Conservative management of benign lesions through ultrasound follow-up or, if symptomatic, minimally invasive surgery, reduces morbidity, preserves fertility, and avoids unnecessary healthcare costs.^4,5 In contrast, women with suspected ovarian cancer benefit from referral to gynecologic oncology centers, where specialized surgical management increases the likelihood of complete tumor removal, improving survival outcomes.^6,7,8

Transvaginal ultrasound is the primary technique used to distinguish benign from malignant ovarian lesions due to its accessibility and high diagnostic accuracy, especially when performed by specialist examiners.^9,10 However, a shortage of experienced examiners, particularly in underserved areas, limits access to timely and accurate diagnoses, leading to delays in treatment and increased risks of unnecessary surgical interventions.^11,12

Advancing Ultrasound Diagnostics with AI Support

The increasing shortage of specialist ultrasound examiners has created significant challenges in the timely and accurate diagnosis of ovarian lesions, resulting in treatment delays and unnecessary surgeries. This highlights the need for innovative approaches to support medical professionals in delivering high-quality care. Our research team, based at the Karolinska Institute and the Science for Life Laboratory (SciLifeLab), in collaboration with KTH Royal Institute of Technology, the Stockholm South General Hospital (Södersjukhuset), and Intelligyn—a company specializing in AI-driven diagnostic tools—is exploring the potential of artificial intelligence (AI) in enhancing the diagnosis of ovarian tumors.

Recent advancements in AI, particularly deep learning, have shown remarkable promise in medical imaging diagnostics, achieving performance comparable to medical specialists in areas such as dermatology and breast cancer screening.^13,14 Building on this progress, our earlier work demonstrated that deep learning models can accurately distinguish between benign and malignant ovarian tumors, matching the diagnostic accuracy of experienced ultrasound examiners.¹⁵

In a recent large-scale international validation study in Nature Medicine, we validated these findings across diverse populations and ultrasound equipment.¹⁶ This study included 3,652 patients from 20 centers in 8 countries and the AI model consistently outperformed 66 ultrasound examiners, including 33 specialists with a median of 17 years of experience. Additionally, a retrospective simulation demonstrated the potential of the AI model to serve as a second independent reader, supporting non-specialist examiners in a clinical triage workflow. In this setup, specialists reviewed only cases of disagreement, leading to a 63% reduction in referrals to specialists and an 18% reduction in incorrect diagnoses. These findings highlight the potential of diagnostic AI support to significantly enhance efficiency and optimize the use of limited healthcare resources.

OV-AID: Discriminating benign from malignant OVarian tumors using computer AIDed diagnostic support

Building on prior advancements in AI-driven diagnostics, the OV-AID Phase I study is a prospective multicenter diagnostic accuracy study designed to evaluate the role of AI in ultrasound-based assessment of ovarian tumors. This study aims to assess the potential of AI to enhance diagnostic precision and complement existing methods.

Study Overview and Objectives

The OV-AID Phase I study investigates the diagnostic accuracy of our previously developed AI model¹⁶  in distinguishing between benign and malignant ovarian lesions. The study compares AI-based assessments to traditional subjective evaluations using pattern recognition¹⁷ and the Assessment of Different NEoplasias in the adneXa (ADNEX)¹⁸ model, including cancer antigen 125 (CA-125),  developed by the International Ovarian Tumor Analysis (IOTA) group. The comparisons will be stratified by the level of expertise of the examining physician (specialists versus non-specialists). The goal is to evaluate how AI can bridge diagnostic gaps and support clinicians in improving patient outcomes.

Study Design

This multicenter observational study is being conducted across 14 centers in five countries: Czech Republic, Italy, Lithuania, Poland, Spain, and Sweden. The histological outcomes from surgery or long-term ultrasound follow-up (at least 9 months) serve as the gold standard for comparison. To maintain objectivity, examiners are blinded to the AI model’s predictions, ensuring that the management of patients remains unaffected.

Eligibility and Recruitment

Eligible participants include women aged 16 and older with newly detected adnexal lesions. Recruitment began in March 2021 with a target enrollment of 700 patients, including at least 400 cases assessed by non-specialist examiners and 300 by specialists. Enrolment will conclude upon reaching the target sample size. Data collection will continue until all participants have confirmed histological outcomes from surgery or, in cases managed conservatively, long-term ultrasound follow-up for a minimum of nine months.

Ethical Approvals

This study has obtained necessary ethical approvals from the Swedish Ethics Review Authority (Etikprövningsmyndigheten) (Dnr 2020-07200, with addenda Dnr 2021-04549, 2021-06367-02, 2023-01834-02, 2023-06127-02, and 2024-04743-02), ensuring compliance with high ethical standards, including informed consent and participant safety.

Looking Ahead

The OV-AID Phase I study is a crucial step toward integrating AI into ovarian tumor diagnostics. Insights gained will guide future studies, including the planned OV-AID Phase II clinical trial, and support the adoption of AI-driven tools to optimize healthcare resources, reduce diagnostic errors, and improve patient outcomes. Findings will be shared in peer-reviewed publications.

Other Ongoing and Planned Studies

In addition to OV-AID Phase I, several other studies are underway or planned to further explore the impact of AI in ovarian tumor diagnostics:

• Retrospective Multi-Reader, Multi-Case (MRMC) Study: This study will involve approximately 30 examiners with different levels of expertise assessing at least 500 ovarian lesions twice, once with and once without AI support, to evaluate the impact of AI support on diagnostic accuracy and inter-observer variability.

• Clinical Feasibility Study (IntelligynAI-FS): This clinical study focuses on the practical integration of AI support into clinical workflows, examining its effect on diagnostic confidence and examiner decision-making. Read more here: ISRCTN90989270.

• The BIO-IMAGE Cohort: A sub-cohort of the OV-AID Phase I study, including at least 200 patients undergoing expert ultrasound assessment at Södersjukhuset and triaged for surgery. This cohort investigates the added value of combining CA-125, circulating tumor DNA (ctDNA), demographic data, and AI-driven image analysis for cancer prediction. The novel multimodal model will be compared with current risk prediction methods, such as subjective assessment by a specialist and the IOTA ADNEX model, to validate its effectiveness in a clinical setting.

• Randomized Controlled Multicenter Study (OV-AID Phase II): Planned to start in 2026, this study will explore the effect of AI support on patient outcomes, management, and cost-benefit measures in ultrasound assessment of ovarian lesions.

Acknowledgments

This study is made possible through the support of the Swedish Research Council (Vetenskapsrådet), the Swedish Cancer Society (Cancerfonden), the Stockholm Regional Council (Region Stockholm), and the Cancer Research Funds of Radiumhemmet (Radiumhemmets Forskningsfonder).

Join Us on Our Journey

For those interested in learning more about OV-AID Phase I and staying updated on our progress, we encourage you to visit our ISRCTN registry entry and follow Intelligyn's website and our forthcoming publications. Together, we can move closer to ensuring that innovative AI solutions meet the needs of both patients and healthcare providers.

References

1. Torre, L. A. et al. Ovarian cancer statistics. CA Cancer J Clin, 68:284–296 (2018). DOI: 10.3322/caac.21456.
2. American College of Obstetricians and Gynecologists. Practice bulletin no. 174: evaluation and management of adnexal masses. Obstet Gynecol, 128:e210–e226 (2016). DOI: 10.1097/aog.0000000000001768.
3. Sharma, A. et al. Risk of epithelial ovarian cancer in asymptomatic women with ultrasound‐detected ovarian masses: a prospective cohort study within the UK collaborative trial of ovarian cancer screening (UKCTOCS). Ultrasound Obstet Gynecol, 40:338–344 (2012). DOI: 10.1002/uog.12270.
4. Yazbek, J. et al. Effect of quality of gynaecological ultrasonography on management of patients with suspected ovarian cancer: a randomised controlled trial. Lancet Oncol, 9:124–131 (2008). DOI:10.1016/s1470-2045(08)70005-6.
5. Froyman, W. et al. Risk of complications in patients with conservatively managed ovarian tumours (IOTA5): a 2-year interim analysis of a multicentre, prospective, cohort study. Lancet Oncol, 20:448–458 (2019). DOI: 10.1016/s1470-2045(18)30837-4.
6. Vernooij, F. et al. Specialized care and survival of ovarian cancer patients in the Netherlands: nationwide cohort study. J Natl Cancer Inst, 100:399–406 (2008). DOI: 10.1093/jnci/djn033.
7. Vergote, I. et al. Prognostic importance of degree of differentiation and cyst rupture in stage I invasive epithelial ovarian carcinoma. Lancet, 357:176–182 (2001). DOI: 10.1016/s0140-6736(00)03590-x.
8. Bristow, R. E. et al. Survival effect of maximal cytoreductive surgery for advanced ovarian carcinoma during the platinum era: a meta-analysis. J Clin Oncol, 41:4065–4076 (2023). DOI: 10.1200/jco.22.02765.
9. Timmerman, D. et al. ESGO/ISUOG/IOTA/ESGE Consensus Statement on pre-operative diagnosis of ovarian tumors. Int J Gynecol Cancer, 31:961–982 (2021). DOI: 10.1136/ijgc-2021-002565.
10. Van Holsbeke, C. et al. Ultrasound methods to distinguish between malignant and benign adnexal masses in the hands of examiners with different levels of experience. Ultrasound Obstet Gynecol,34:454–461 (2009). DOI: 10.1002/uog.6443.
11. Van Holsbeke, C. et al. Ultrasound experience substantially impacts on diagnostic performance and confidence when adnexal masses are classified using pattern recognition. Gynecol Obstet Invest,69:160–168 (2010). DOI: 10.1159/000265012.
12. Timmerman, D. et al. Subjective assessment of adnexal masses with the use of ultrasonography: an analysis of interobserver variability and experience. Ultrasound Obstet Gynecol, 13:11–16 (1999). DOI: 10.1046/j.1469-0705.1999.13010011.x.
13. Esteva, A. et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542:115–118 (2017). DOI: 10.1038/nature21056.
14. McKinney, S.M. et al. International evaluation of an AI system for breast cancer screening. Nature,577:89–94 (2020). DOI: 10.1038/s41586-019-1799-6.
15. Christiansen, F. et al. Ultrasound image analysis using deep neural networks for discriminating between benign and malignant ovarian tumors: comparison with expert subjective assessment. Ultrasound Obstet Gynecol, 57:155–163 (2021). DOI: 10.1002/uog.23530.
16. Christiansen, F. et al. International multicenter validation of AI-driven ultrasound detection of ovarian cancer. Nat Med, 31:189–196 (2025). DOI: 10.1038/s41591-024-03329-4.
17. Van Calster, B. et al. Discrimination between benign and malignant adnexal masses by specialist ultrasound examination versus serum CA-125. J Natl Cancer Inst, 99:1706–1714 (2007). DOI: 10.1093/jnci/djm199.
18. Van Calster, B., et al. Evaluating the risk of ovarian cancer before surgery using the ADNEX model to differentiate between benign, borderline, early and advanced stage invasive, and secondary metastatic tumours: prospective multicentre diagnostic study. BMJ, 349:g5920 (2014). DOI: 10.1136/bmj.g5920.