FläKI
Widespread use of AI models by Saarland ophthalmologists to improve medical care, increase the acceptance of AI among physicians and patients, and promote innovative AI services in ophthalmology.
The research project FläKI aims to transform ophthalmology (eye care) through the widespread use of artificial intelligence (AI) in medical facilities. The first phase of the project focuses on the development of an AI-compatible data space consisting of high-quality annotated ophthalmology data for the data protection-compliant training of AI models. Second, a feedback system (Human-in-the-Loop, HITL) will be implemented. The HITL system enables medical professionals to verify the results of the AI models and feed the feedback back into the system, enabling continuous AI learning. The widespread use of AI models is to be achieved with the help of the online portal "VISYOnet Qualitätsnetzwerk Saar," as this is already used by 98% of Saarland ophthalmologists.
FläKI The innovative data space enables the development of new AI models for ophthalmology as well as the continuous verification of AI results. The generic design of the developed infrastructure will also enable other medical institutions to easily connect innovative AI services in the future. FläKI a contribution to improving the quality of treatment and increasing the acceptance of AI among doctors and patients.
Our role in the project
The August-Wilhelm Scheer The institute supports the development and implementation of the data protection strategy. Drawing on its expertise in the field of artificial intelligence, the institute is responsible for the design and development of the human-in-the-loop system, the multi-AI adapter, and an AI clustering model.
Privacy Policy
Task of the August-Wilhelm Scheer The Institute's mission is to support the creation of the data protection concept and its implementation within the HITL system.
Result: GDPR-compliant data protection concept as well as ensuring data security/GDPR compliance
Basics of a participation-oriented data governance strategy
Through its expertise in the field of artificial intelligence, the August-Wilhelm Scheer The institute provides valuable expertise on the technical design of a data room, as well as on the structuring and efficient processing of data. The institute participates in expert roundtables organized by K8 to gather direct feedback for the technical development and is also available as a contact and discussion partner for technical details of the HITL system. The institute also supports the development of the results for the creation of the data governance strategy.
Result: Short documentation of the results of the expert rounds (overview of recommendations for the design of the data room)
Development of the multi-AI adapter and cluster process
The August-Wilhelm Scheer The institute is responsible for the technical design and development of the multi-AI adapter for connecting AI systems to the HITL system based on existing requirements. Furthermore, the institute is responsible for the technical design of suitable AI methods based on the defined requirements, as well as the training and iterative development of an AI model for clustering based on OCT scans, anonymized patient data, and clinical information, taking into account the defined target parameters.
Results: Multi-AI adapter, AI clustering component
Interface development
The August-Wilhelm Scheer The institute is responsible for defining and developing internal interfaces for data transfer as well as for defining and developing interfaces for external data transfer.
Result: Development and integration of internal and external interfaces for data transfer
Development and integration of the HITL system
The August-Wilhelm Scheer The institute is responsible for creating prototype dashboards and mockups for conducting user-centered feedback loops with the relevant target groups. The institute is also responsible for the design and iterative development of the human-in-the-loop system. Based on the recorded requirements for the user experience and user interface, a dashboard is being developed that the domain experts can use to verify the results of the AI-based clustering, and the results are then fed back into the system.
Results: HITL system, increasing societal trust in intelligent systems
Evaluation and demonstration
The August-Wilhelm Scheer The institute supports the coordination and implementation of the preparation and actively participates in the real-world tests and the final event. It also supports the organization, planning, and implementation of the real-world tests and the final event, and incorporates the resulting findings into the AI solutions.
Results: Finalized project solution and demonstration
The initial situation
Overview of the need for action
In Germany, only 7% of clinical organizations use AI, despite extensive data available in the medical field. This is due to the lack of structuring and annotation of the data. Legal guidelines, technical barriers and lack of acceptance hinder the widespread use of AI in the healthcare sector, including eye care. Data silos are emerging in ophthalmology because large amounts of data, particularly optical coherence tomography (OCT) scans, are not comprehensively stored or made available. Although AI models are successful in medicine, they encounter obstacles in ophthalmology such as lack of trust, validity problems and difficulties in integrating into everyday clinical practice.
only 7% of clinical organizations in Germany use AI
Amounts of data exist, but have not yet been usable due to structural obstacles
Your contact person
FläKI
Development and testing of a comprehensive human-in-the-loop AI system using the example of ophthalmological treatment (Phase 1)
Daniela Podevin
daniela.podevin@aws-institut.de
+ 49 152 34643 509
Your contact person
FläKI
Development and testing of a comprehensive human-in-the-loop AI system using the example of ophthalmological treatment (Phase 1)
Daniela Podevin
daniela.podevin@aws-institut.de
+ 49 152 34643 509
Our solution approach in focus
In the first project phase, a comprehensive data room is set up, which consists of: manufacturer-dependent and manufacturer-independent data (OCT data, visual acuity values, eye pressure data, fundus findings, patient information, clinical data). The integration of a HITL system enables the verification of the AI results by specialists throughout the entire development process. Internationally recognized standards such as HL7/FHIR and Direct/DirectTrust are taken into account for data exchange. The data space developed will also be used for future AI models in ophthalmology.
The integration of a multi-AI adapter (MAA) as an interface between AI algorithms and the VISYOnet online portal plays a central role. The MAA enables the integration of various AI models and thus allows a low-threshold expansion of functionality.
As an example, scalable clustering based on the data space is implemented to perform patient categorization. This gives ophthalmologists access to similar medical cases and promotes the exchange of successful treatment information within the specialist community. The technical components and interfaces are validated in an end-to-end test, whereby the integration capability, usability and user experience of the HITL system are also checked.
The data strategy and interfaces of the data room are continuously aligned with the Health.AI Dataspace to ensure seamless integration with other projects. Incorporating international standards for data management and data exchange ensures interoperability and enables the creation of a flexible HITL-AI adapter in the healthcare sector. The integration of health insurance companies as cooperation partners ensures that the desired quality improvements in patient treatment are implemented effectively.
In later phases, complex AI services, including those from the project Ophthalmo-AI, be introduced.
Funding notice
The Project FläKI is funded by the Federal Ministry of Research, Technology and Space.
Funding code: 03WIR5607B
Running time: 01.01.2024 – 31.10.2025
