Periodic Reporting for period 1 - I-SCREEN (I(eye)-SCREEN: A real-world AI-based infrastructure for screening and prediction of progression in age-related macular degeneration (AMD) providing accessible shared care)
Reporting period: 2024-01-01 to 2024-12-31
The aim of I-SCREEN is to develop an artificial intelligence (AI)-based diagnostic decision support system for the screening and monitoring of age-related macular degeneration (AMD), a leading cause of vision loss. The multidisciplinary project brings together clinical retina experts, eye-care specialists working in community-based settings, as well as computer scientists working at the cutting edge of AI development. To develop an innovative and trustworthy AI support system, I-SCREEN incorporates three clinical studies for the acquisition of high-quality data using optical coherence tomography (OCT) imaging, a non-invasive imaging modality that provides a detailed 3D characterization of the retina. Clinical studies (WP1) are carried out at 7 clinics and a community-based screening (WP4) will take place at 21 optometry sites across Europe. A small-medium sized enterprise is supporting the project by providing a cloud-based platform for joint data collection and implementation of AI in community-based settings.
I-SCREEN has 4 main objectives:
Objective 1: Invent a methodology of AI for retinal image interpretation from longitudinal volumetric OCT scans addressing the challenges of high dimensional data, temporality and data efficient training, as well as model interpretability required to develop trustworthy AI-based predictive models of AMD progression. This will be achieved within collaboration between WP1 and WP2.
Objective 2: Run large prospective observational studies of natural observation of patients with intermediate AMD (SUDETES) and functional atrophic AMD (APENNINES) addressing the challenge of limited longitudinal data availability and currently limited evaluation of the performance of the developed AI-based predictive models. This is the focus of WP1.
Objective 3: Automated diagnosis of intermediate AMD on low-cost devices available in community-based opticians’/optometrists’ offices addressing the challenge of transferring the models from the clinical setting to a community-based scenario applicable to AMD patients in their natural environment (PYRENEES). This will be achieved through collaboration between WP3 and WP4.
Objective 4: Predictive model of AMD disease progression using OCT imaging applicable to both high-end and low-cost OCT, addressing the challenges of image domain shift, and patient as well as disease heterogeneity, by providing a personalized risk estimator of future AMD progression and therapeutic response. This objective will be mainly worked on in WP2.
I-SCREEN has 4 main objectives:
Objective 1: Invent a methodology of AI for retinal image interpretation from longitudinal volumetric OCT scans addressing the challenges of high dimensional data, temporality and data efficient training, as well as model interpretability required to develop trustworthy AI-based predictive models of AMD progression. This will be achieved within collaboration between WP1 and WP2.
Objective 2: Run large prospective observational studies of natural observation of patients with intermediate AMD (SUDETES) and functional atrophic AMD (APENNINES) addressing the challenge of limited longitudinal data availability and currently limited evaluation of the performance of the developed AI-based predictive models. This is the focus of WP1.
Objective 3: Automated diagnosis of intermediate AMD on low-cost devices available in community-based opticians’/optometrists’ offices addressing the challenge of transferring the models from the clinical setting to a community-based scenario applicable to AMD patients in their natural environment (PYRENEES). This will be achieved through collaboration between WP3 and WP4.
Objective 4: Predictive model of AMD disease progression using OCT imaging applicable to both high-end and low-cost OCT, addressing the challenges of image domain shift, and patient as well as disease heterogeneity, by providing a personalized risk estimator of future AMD progression and therapeutic response. This objective will be mainly worked on in WP2.
I-SCREEN started in January 2024. During the first 12 months of the project, major effort was put into the following activities to achieve the defined objectives:
Objective 1 - During the first 12 months of the I-SCREEN project, significant focus was placed on the design of clinical studies aimed at generating data for the development of advanced AI methodologies. The clinical studies (WP1), SUDETES and APENNINES, have been strategically designed to enable a comprehensive evaluation of both local and global changes associated with AMD. The study protocols, developed in collaboration with AI teams (WP2 and WP3), establish a robust foundation for advancing AI-driven innovations in diagnostic and predictive modelling. This collaborative approach ensures that the AI tools developed are both clinically relevant and technically sophisticated, paving the way for meaningful advancements in the field.
An initial approach to advancing the capabilities of AI models has been started as part of efforts in WP2. The methodological advancement has focused on improving data efficiency of AI algorithms, for which a new method utilizing self-supervised learning has been developed, that learns from a pair of consecutive scans, and which has been shown to be very suitable for capturing hidden temporal factors in a longitudinal dataset.
Objective 2 - With regards to Objective 2, during RP1, the primary focus was on establishing the necessary infrastructure for conducting large-scale prospective studies across all seven clinical sites. This included obtaining ethics approvals for the SUDETES and APENNINES studies, securing legal clearances, and completing the technical setup required to initiate the studies. Notably, these studies are designed to incorporate MDR-certified AI tools for precise monitoring and quantification of disease progression—a groundbreaking approach for many clinics. Significant efforts were dedicated to finalizing the legal agreements and frameworks associated with these arrangements.
Objective 3 - During RP1 of I-SCREEN, a robust network of community-based opticians and optometrists was established in collaboration with the seven clinical sites (WP4). This network now comprises 21 optician and optometrist offices, which will play a critical role in collecting community-based data. This data will be instrumental in developing an AI system capable of automated diagnosis of intermediate AMD using low-cost imaging devices. In the long term, this network will also facilitate the implementation of AI-based diagnostic solutions in community healthcare settings. Important to highlight are the different patient pathways across Europe, where the collaboration between clinics and optometry or optician sites varies greatly from country to country.
In addition to building this network, significant progress was made in developing the infrastructure necessary for community-based diagnostic software. Within WP3, a data collection platform was successfully developed, which will serve as the cornerstone of AI-driven screening initiatives. The platform supports the upload and management of clinical and imaging data.
Objective 4 - During RP1, in the absence of imaging data from the clinical studies, we have started conceptualizing the predictive model that will be developed for AMD progression. In addition, we have worked on developing necessary tools and methods that will be required to later compare the high-end and low-end OCTs. For both of these an approach for image registration, i.e. the algorithm for spatial alignment of two OCT scans acquired either from the same patient at two different time-points or on the same day with two different devices, a characteristic property of both SUDETES and APENNINES, is being developed. The approach allows to have a data-efficient and well-regularized spatial transformation estimate that maps the en-face views of two OCT scans.
Objective 1 - During the first 12 months of the I-SCREEN project, significant focus was placed on the design of clinical studies aimed at generating data for the development of advanced AI methodologies. The clinical studies (WP1), SUDETES and APENNINES, have been strategically designed to enable a comprehensive evaluation of both local and global changes associated with AMD. The study protocols, developed in collaboration with AI teams (WP2 and WP3), establish a robust foundation for advancing AI-driven innovations in diagnostic and predictive modelling. This collaborative approach ensures that the AI tools developed are both clinically relevant and technically sophisticated, paving the way for meaningful advancements in the field.
An initial approach to advancing the capabilities of AI models has been started as part of efforts in WP2. The methodological advancement has focused on improving data efficiency of AI algorithms, for which a new method utilizing self-supervised learning has been developed, that learns from a pair of consecutive scans, and which has been shown to be very suitable for capturing hidden temporal factors in a longitudinal dataset.
Objective 2 - With regards to Objective 2, during RP1, the primary focus was on establishing the necessary infrastructure for conducting large-scale prospective studies across all seven clinical sites. This included obtaining ethics approvals for the SUDETES and APENNINES studies, securing legal clearances, and completing the technical setup required to initiate the studies. Notably, these studies are designed to incorporate MDR-certified AI tools for precise monitoring and quantification of disease progression—a groundbreaking approach for many clinics. Significant efforts were dedicated to finalizing the legal agreements and frameworks associated with these arrangements.
Objective 3 - During RP1 of I-SCREEN, a robust network of community-based opticians and optometrists was established in collaboration with the seven clinical sites (WP4). This network now comprises 21 optician and optometrist offices, which will play a critical role in collecting community-based data. This data will be instrumental in developing an AI system capable of automated diagnosis of intermediate AMD using low-cost imaging devices. In the long term, this network will also facilitate the implementation of AI-based diagnostic solutions in community healthcare settings. Important to highlight are the different patient pathways across Europe, where the collaboration between clinics and optometry or optician sites varies greatly from country to country.
In addition to building this network, significant progress was made in developing the infrastructure necessary for community-based diagnostic software. Within WP3, a data collection platform was successfully developed, which will serve as the cornerstone of AI-driven screening initiatives. The platform supports the upload and management of clinical and imaging data.
Objective 4 - During RP1, in the absence of imaging data from the clinical studies, we have started conceptualizing the predictive model that will be developed for AMD progression. In addition, we have worked on developing necessary tools and methods that will be required to later compare the high-end and low-end OCTs. For both of these an approach for image registration, i.e. the algorithm for spatial alignment of two OCT scans acquired either from the same patient at two different time-points or on the same day with two different devices, a characteristic property of both SUDETES and APENNINES, is being developed. The approach allows to have a data-efficient and well-regularized spatial transformation estimate that maps the en-face views of two OCT scans.
Following results have been achieved within the first reporting period:
- A network and infrastructure consisting of 7 clinical centers and 21 optometry sites has been set up within WP4. This network is a first of its kind and will work together to facilitate the screening of the population, in order to help detect AMD at an early, non-vision compromising stage.
- A cloud-based platform for joint data collection and sharing within the consortium has been developed within WP3. The platform allows for the safe storage of imaging data as well as clinical data of all patients who have consented to participating in the clinical studies. The platform also enables a telemedicine doctor (board-certified ophthalmologist) to access uploaded OCT images in order to give a timely referral recommendation to the optometrist and patient.
- A network and infrastructure consisting of 7 clinical centers and 21 optometry sites has been set up within WP4. This network is a first of its kind and will work together to facilitate the screening of the population, in order to help detect AMD at an early, non-vision compromising stage.
- A cloud-based platform for joint data collection and sharing within the consortium has been developed within WP3. The platform allows for the safe storage of imaging data as well as clinical data of all patients who have consented to participating in the clinical studies. The platform also enables a telemedicine doctor (board-certified ophthalmologist) to access uploaded OCT images in order to give a timely referral recommendation to the optometrist and patient.