Diagnosing endometriosis and adenomyosis with AI

Endometriosis and adenomyosis affect the health of over 190 million individuals assigned female at birth (AFAB). Endometriosis is a chronic, systemic, inflammatory disease characterised by the presence of endometrium-like cells (i.e. similar to the uterus lining) outside of the uterus cavity. Whereas adenomyosis is a chronic, inflammatory disease characterised by the presence of endometrium-like cells within the uterus wall. Both conditions can cause a wide range of symptoms, like heavy periods, chronic fatigue or infertility.

Currently, diagnosis depends on healthcare professionals analysing transvaginal ultrasounds, the first-line imaging modality across the Netherlands and Europe. While AI has significant potential to automate the analysis of these images, its effectiveness depends on high-quality data. That’s why project lead Bianca Schor and her team in Amsterdam UMC set out to create a large and diverse dataset that will be made available as open-source data. In the second part of the project, several AI-models will be developed into a prototype using the dataset.

AI4EA is supported by the first Dutch consortium that focuses on AI research for endometriosis and adenomyosis. Coordinated by Amsterdam UMC, the consortium brings together 10 hospitals that specialise in endometriosis and adenomyosis and the patient foundation ‘Endometriose Stichting’. All 10 hospitals that are part of the Dutch consortium will be contributing to the open-source dataset.

To ensure that the data collected for the dataset is representative and meets all regulations, the project group follows a standardised protocol that was designed specifically for this project. 'A step-by-step method ensures that data collection is reproducible and comparable across other project groups. With the protocol, we also aim to help bridge the gender data gap, which is still very large', Bianca says. Another point is reducing bias in the data. ‘The bias can be multilevel. It is not only introduced by having a one-sided representation of the population, but also by differences in which machine is used, or the technique used and level of expertise in the ultrasound examination. That’s why we work with different hospitals and take additional measures before running the data through the AI models.’

The dataset will serve as the foundation for several AI models that can support healthcare professionals in identifying signs of endometriosis and adenomyosis on ultrasound scans. Over time, the models will perform increasingly complex tasks, ultimately enabling them to identify abnormalities that may support diagnosis. However, a medical professional will always retain the final responsibility for diagnosis and decision-making, for both medical and legal reasons.

Bianca believes complex medical issues, such as endometriosis and adenomyosis, require technology and innovation to make progress. She says: 'After showing great success in the field of radiology, AI should be explored throughout healthcare. At the very least, professionals in the medical field need to be more aware of both the risks and advantages of AI.’

The successful implementation of AI in clinical practice remains a challenge for gynaelogical imaging. As Bianca describes, many promising AI models fail to deliver real-world impact because they are developed without sufficient involvement from the professionals and patients who will ultimately use them. To address this challenge, the consortium actively involves gynaecologists, patients and other stakeholders throughout the development process. Their perspectives and different areas of expertise help ensure that the developed AI models address real clinical needs and can be integrated into clinical practice more effectively.

Bianca highlights that this project is a first step towards improving care for patients with endometriosis and adenomyosis. She says: ‘Women's health is still considered a niche, even though it affects half of the world’s population. That’s why we are grateful to ZonMw for providing a platform for our project.’ She hopes the impact of AI4EA extends beyond the development of AI models. By making the dataset, protocol and publications openly available, the consortium enables researchers worldwide to accelerate progress in understanding endometriosis, adenomyosis and women’s health more broadly.