Can ovarian cancer detection be improved using AI-driven diagnostic support applied to ultrasound images?

Plain English Summary

Background and study aims
Ovarian tumors are common and their management is dependent on the risk of malignancy. Patients with malignant ovarian tumors benefit from being referred to a gynecologic oncologist for debulking surgery, indicating higher survival rates. On the other hand, benign cysts should be handled conservatively with ultrasound follow-up or with minimally invasive surgery, while preserving fertility and avoiding unnecessary suffering. Transvaginal ultrasound assessment has a central role in the diagnostics of ovarian tumors as it has high accuracy, at least in the hands of ultrasound experts. There is currently a shortage of sonographers with enough experience though, as malignant tumors are relatively uncommon. Thus, there is a need to find alternatives to improve the diagnostics of ovarian tumors. In other medical fields, researchers have been able to develop automated imaging tools to improve diagnostic accuracy, for example when searching for breast cancer and bone fractures. Recent advances in computerized image analysis have been powered by deep neural networks (DNNs), which are computational versions mimicking the biological nervous system. Recently, researchers have developed and validated a DNN model (Ovry-Dx) to discriminate benign from malignant ovarian tumors based on 3077 ultrasound images from 758 patients. Ovry-Dx had a diagnostic accuracy comparable to human expert assessment. Further validating studies are needed to explore the generalizability and robustness of DNN models, and the clinical benefits in the hands of less experienced examiners. This study aims to compare the accuracy in differentiating benign from malignant masses using DNN models as compared to subjective assessment using pattern recognition or the IOTA-ADNEX model, by examiners of different levels of expertise.

Who can participate?
Women aged 15 years and over with a recently detected ovarian tumor, planned for either surgery or long time (at least 21 months) follow-up

What does the study involve?
Subjective assessment using pattern recognition and the IOTA-ADNEX model score will be compared to the DNN-model assessment. The histological outcome from surgery or long-time ultrasound follow-up (minimum 21 months) serves as the gold standard.

What are the possible benefits and risks of participating?
There are neither any benefits nor any risks as the DNN model outcome will not be shown to the examiner. The images will not be analysed by the DNN model until all patients have been examined. Therefore, participation will not affect the management of the patient.

Where is the study run from?
Karolinska Institutet (Sweden)

When is the study starting and how long is it expected to run for?
January 2020 to June 2026

Who is funding the study?
1. Swedish Research Council (Sweden)
2. Swedish Cancer Society (Sweden)
3. Stockholms Läns Landsting (Sweden)
4. Radiumhemmets Forskningsfonder (Sweden)

Who is the main contact?
Elisabeth Epstein, elisabeth.epstein@ki.se

Contact information

Prof Elisabeth Epstein
Principal Investigator

Sjukhusbacken 10
Stockholm
11883
Sweden

ORCID ID	0000-0003-2298-7785
Phone	+46 (0)706699019
Email	elisabeth.epstein@ki.se

Study information

Eligibility

Locations

Countries of recruitment

• Czech Republic
• Italy
• Lithuania
• Philippines
• Poland
• Spain
• Sweden

Study participating centres

Sponsor information

Karolinska Institute
University/education

Sjukhusbacken 10
Stockholm
11883
Sweden

Phone	+46 (0)8 524 800 00
Email	annsofie.wall@ki.se
Website	https://ki.se/kisos
"ROR"	https://ror.org/056d84691

Funders

Funder type

Government

Results and Publications

Study outputs

Additional files

ISRCTN88222986 Prospective_Study_Protocol_v1.pdf

ISRCTN88222986_PROTOCOL_V2.pdf

Editorial Notes

28/05/2024: The following changes were made to the study record:
1. Uploaded protocol.
2. The acronym was changed from 'OV-AID' to 'OV-AID, Phase I'.
3. Ethics approval details added.
4. The target number of participants was changed from 'Minimum 600 participants, where at least 150 tumors are malignant, and at least 200 assessments are performed by non-experts' to 'Minimum 700 participants, and at least 400 assessments are performed by non-experts'.
5. Endogyn, Institute for the Care of Mother and Child, I.R.C.C.S. Burlo Garofolo Institute for Maternal and Child Health, Department of Obstetrics and Gynecology, IRCCS San Gerardo Dei Tintori, Department of Medicine and Surgery, University of Milano-Bicocca, Department of Obstetrics and Gynecology Fondazione Poliambulanza Istituto Ospedaliero, Obstetrics and Gynecology Unit, Forlì and Faenza Hospitals, Gynecology and Breast Care Center, Mater Olbia Hospital, Medical University of Lublin, Medical University of Silesia, Department of Obstetrics and Gynecology, Rizal Medical Center, Department of Obstetrics, Gynecology, and Reproduction, Hospital Universitario Dexeus, Department of Obstetrics and Gynaecology, Lithuanian University of Health Sciences were added to the study participating centres.
6. The interventions were updated.
7. Czech Republic, Italy, Poland, Philippines, Spain and Lithuania were added to the countries of recruitment.
06/11/2023: Uploaded protocol (not peer-reviewed) as an additional file.
8. Radiumhemmets Forskningsfonder was added to the funders.
9. The intention to publish date was changed from 01/03/2026 to 30/06/2026.
31/03/2023: Trial's existence confirmed by Swedish Cancer Society (Cancerfonden).