A successful story

Digital transformation in health: Application of artificial intelligence in the reading of mammographic findings in the Health Center of Primorje-Gorski Kotar County

Image by Pete Linforth from Pixabay

Challenges

The health center of Primorje-Gorski Kotar County, as well as many other health institutions in Croatia and beyond, faces long and resource-intensive processes of reading mammograms. One of the key challenges is the chronic lack of radiologists, resulting in long waiting lists, delayed diagnosis, and an increased risk of missing early signs of breast cancer.

In the context of early detection of breast cancer, time is a key factor. Each day of delay in reading can mean the difference between a successful treatment and a late, more complicated intervention. In addition, the interpretation of mammograms often involves a certain dose of subjectivity, which further increases the pressure on the limited number of available radiologists and reduces diagnostic consistency among specialists.

Due to all of the above, it has become necessary to consider the use of artificial intelligence (AI) as a technological partner in addressing these challenges. AI would not replace professionals, but allow them to do their job more efficiently, quickly and safely.

Solution

As part of the Test Before Invest (TBI) activities, EDIH Adria proposed and tested the application of an AI model for automatic detection and localisation of lesions on mammograms, based on the YOLOv11 algorithm – one of the most advanced and fastest computer vision models for real-time image analysis.

Detection and localisation of suspected lesions - YOLOv11

The YOLOv11 (You Only Look Once) model uses deep convolutional neural networks to analyze the entire image in one pass, allowing very fast and precise detection of suspicious masses. The system detects and classifies lesions as potentially benign or malignant and automatically indicates their positions in the mammogram picture, along with the level of confidence for each finding.

During testing, the model showed significant results in terms of speed and accuracy, and an additional advantage is that it can be adjusted and additionally trained on domestic data, which would increase the specificity in the context of the Croatian health system.

The system is conceived as a tool that supports patient triage – it indicates risky cases and allows for their priority examination, thus significantly contributing to the optimization of the diagnostic process in conditions of limited resources.

Results and Benefits

The test phase of the solution confirmed the following benefits:

✅Significant acceleration in the processing of findings – detection only takes a few seconds per shot
✅Automatic case prioritisation – the system flags the most critical cases for a quick overview
✅Reducing the cognitive load of radiologists – routine findings can be quickly filtered, leaving more time for complex cases
✅Standardisation of readings – reduces variability between different professionals
✅Preparedness for integration into clinical practice – interface and model can be connected to existing hospital information systems (PACS, RIS, etc.)
✅Scalability – the solution can be extended to other institutions facing similar challenges

Lessons learned

The project highlighted several key findings:

  • Data quality – successful implementation of AI systems requires clearly labelled, high-quality and diverse data
  • User involvement – Collaboration with medical staff during testing was key to design a solution that fits real needs
  • Flexible solution architecture – allowing rapid adaptation to new clinical conditions and regions
  • Trust in the system – transparent decision-making logic and visualisation of results increase acceptance among radiologists

Conclusion

Given the challenges that the health system is increasingly facing, AI is becoming a necessary ally in maintaining the quality and speed of diagnostics. By introducing a system based on the YOLOv11 model, it is possible to implement a concrete, efficient and scalable AI solution that meets the challenges of everyday life in public health.

This solution not only contributes to the early detection of breast cancer, but also serves as an example of good practice for other health institutions that want to modernise their diagnostic processes without major investments and with minimal changes to existing workflows.

After the testing phase of the mentioned solution, it is necessary to define all the steps necessary for the procurement of services and resources necessary for the realization of the project, which includes, among other things, the development of technical specifications of the system. Also, in agreement with other project stakeholders, a financial plan and cost statement will be drawn up with the aim of finding sources of financing that would enable the start of the operational phase of the project and the final commissioning. It is important to note that prior to full implementation, the system will be trained on a large amount of findings and data from the database of the Health Center of the Primorje-Gorski Kotar County in order to finally verify its accuracy and effectiveness and to confirm all the results and benefits identified in the preliminary phase.

Facebook
Twitter
LinkedIn

Read more

Read more

what is

de minimis?

Low-value aid; the total amount of which may not exceed €200,000 per undertaking, or €100,000 in the case of an undertaking engaged in road transport for hire or reward, in any period within three fiscal years.

In doing so, all de minimis aid shall be taken into account (aggregated) irrespective of the instrument, purpose and level of the de minimis granting authority.

EDIH Adria

Log in

We will use the personal data collected by this application in accordance with Privacy Policy.

Edih Adria
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.