The aim is to provide physicians with a more objective tool for monitoring changes over time and to help improve the accuracy of screening for retinopathy of prematurity.
Retinopathy of prematurity (ROP) is a serious vascular disease affecting the immature retina of preterm infants and can lead to severe visual impairment or even blindness. One of the challenges of current screening is the subjective nature of image assessment. Retinal haemorrhages can vary considerably in appearance, may have low contrast, may be partially resorbed, or may be difficult to identify.
The project, AI System for Longitudinal Monitoring and Quantification of Retinal Haemorrhages to Improve the Objectivity of Retinopathy of Prematurity Screening, will focus specifically on their automated detection, measurement and long-term monitoring.
“Our aim is to provide physicians with an objective quantitative tool that automatically identifies haemorrhages, measures their extent and other characteristics, and shows how the findings change over time. The goal is not to replace ophthalmologists with artificial intelligence, but to provide them with an additional objective source of information to support their decision-making,” explains Jan Kubíček from the Faculty of Electrical Engineering and Computer Science at VSB-TUO.
The research builds on activities carried out within the LERCO project, under which a team from FEI and the Department of Ophthalmology at University Hospital Ostrava developed and pilot-tested the first AI algorithms for the automated detection and analysis of retinal haemorrhages. Neoli Research funding will enable the team to advance towards a more robust solution capable of working with different types of retinal imaging data and, importantly, with repeated examinations of the same patient.
The project is expected to result in the first experimental version of a clinical software environment. Using AI, the system should be able to detect and precisely delineate areas of haemorrhage in retinal images and subsequently measure parameters such as the number and area of lesions, their size, shape and spatial distribution.
“A key feature is the ability to compare repeated examinations of an individual child. This will provide physicians with information on whether the pathological findings are worsening, remaining stable or, conversely, regressing between follow-up examinations. We also intend to use the data to model the progression of haemorrhages and, in the longer term, to predict their further development,” adds Kubíček.
Artificial intelligence, however, will not make decisions instead of the physician. Its role is to help accurately quantify changes in the retina and show how the findings in an individual child develop between examinations. This will provide physicians with an additional objective basis for their clinical decision-making.
In the long term, this approach could contribute to the standardisation of ROP screening, more accurate monitoring of the progression or regression of pathological changes, and better timing of follow-up examinations.
Collaboration with University Hospital Ostrava
The project directly builds on the existing collaboration between VSB-TUO and University Hospital Ostrava. The researchers will work with clinical images acquired using the Clarity RetCam 3 and Phoenix ICON systems. The inclusion of the Phoenix ICON system represents an important step forward, as it is currently used for ROP screening at University Hospital Ostrava.
The research involves physicians Juraj Timkovič and Kristýna Maršolková from the Department of Ophthalmology at University Hospital Ostrava, who provide clinical consultations and expert image annotations. These annotations serve as reference data from which the AI learns to identify pathological areas.
The FEI research team consists of Jan Kubíček, Avinash Bansal and Marie Chadrabová, who is also focusing on AI-based analysis of retinal haemorrhages in her doctoral research.
Seven Projects Awarded Funding
In the pilot year of the Neoli Research programme, the Neoli Foundation selected seven research projects for funding, allocating a total of CZK 2.784 million. The projects focus on research into preterm birth, approaches to its diagnosis and prevention, as well as new methods and technologies in the care of preterm infants.
“I greatly appreciate the support from Neoli Research, as it allows us to build on the results achieved within LERCO and move them closer to real-world clinical application. I also see great value in combining our technological expertise in AI and medical image processing with a specific clinical need. It is precisely this combination that creates the best conditions for ensuring that the outcome is not merely another AI model, but gradually develops into a practical tool that can support physicians in the care of preterm infants,” concludes Jan Kubíček.