Practical example and current challenges
- Is the result of the removed tissue positive or negative? This is the key question for many patients fearing a cancer diagnosis. In order to enable a precise therapy, genetic characteristics of cells are visually marked with colourings. In doing so, it is important to determine the proportion of cells or tissue components that have been marked in this way. This so-called marker quantification is an essential part of the histopathological diagnosis by a pathologist and a foundation of the therapy.
- The mostly automated marker quantification offers the potential for an increased use of artificial intelligence in pathology, as it leads to a more reliable and faster interpretation of the results. One problem, however, arises from the use of different markers and digitization systems.
- Therefore, the launch of AI-supported applications for marker quantification usually fails due to the variability of on-site conditions and a lack of validation and certification of these applications. One solution is to build up a broad data pool, which can form the basis for both the development of solutions and their validation. Digitised histological sections and the potential for broad clinical application are critical success factors for the development of an application.
- A further challenge consists in the present lack of transparency and comparability of solutions, which makes it difficult to assure quality on a broad basis. By using a common infrastructure for data, applications, and commercial services, it is possible to benchmark results and enhance the system. Thus, comprehensive quality assurance can be attained.
- As an important element of this solution, a data marketplace provides data for the development of AI applications for marker quantification in a transparent manner, enables the labelling of data sets and benchmarking, and ensures access to the models.
What added value does the "GAIA-X project" offer?
- GAIA-X enables secure data storage in the cloud and establishes an authorization and security mechanism to separate training and validation data and to make them independently accessible - furthermore, transparent, anonymized data storage for benchmarking is supported. Developers, service providers and users can thus work together in independent teams to continuously improve solutions.
- The integration of cloud storage in Germany or the EU as GAIA-X nodes allows decentralized and compliant application of AI algorithms, exactly where the sensitive data is located. GAIA-X will connect edge nodes for large users. On-site quality assurance can be ideally complemented by cross-institutional benchmarking.
- Prof. Dr. Peter Hufnagl, Charité – University Medicine Berlin, project coordinator for EMPAIA.org
- Prof. Dr. Sahin Albayrak – TU Berlin, DAI-Laboratory
- Prof. Dr. Horst Hahn – Fraunhofer MEVIS
- Thomas Pilz – Quality assurance initiative Pathology QuIP GmbH