The Department of Medical Informatics at KonKIS24
Four AI service centers were created as part of the governmental strategy “AI made in Germany”. Each one of them covers a main area: KISKKI for medicine and energy, Hessian AI, West AI, and KI-Service center Berlin Brandenburg. KonKIS 24 was held in Göttingen on the 18th and 19th of September and was the first conference of the four centers. The program offered panel discussions, demonstrations, posters, workshops, and scientific sessions around AI's present and future in Germany.
Our institute participated in three activities:
A panel discussion: “AI use in healthcare: current technical, legal and ethical developments (AI Act), challenges and opportunities,” moderated by Prof. Dr. Dagmar Krefting (Director of the Institute for Medical Informatics at the University Medical Center Göttingen). Top-class make-changers panelists included Dr. Larisa Wewetzer (Ottobock, global development for digital health Solutions), Dr. Malte Schmieding (Federal Ministry of Health, Artificial Intelligence in Unit 511 New Technologies and Data Use), PD Dr. Jana Zschüntzsch (University Medical Center Göttingen, rare diseases), Dr. Udo Schneider (Techniker Krankenkasse, Department of Supply Management), and Prof. Dr. Helena Zacharias (Hannover Medical School, clinical data science). The dialogue shows the inside of current applications, letting us know how, for example, Ottobock will create AI models with data from patients and let them learn from every new patient that uses the product, improving the model and adjusting it when required, of course, following the regulations for each country in which their products are using the Azure ML solutions. Another exciting aspect is that in the future, all projects funded by the Ministry of Health on basic research should collect and prepare the data to allow primary and secondary usage for developing AI solutions; until now, not all basic research projects have considered this aspect. In general, there was a consensus that investments in infrastructure and the digitalization process to enable the research and development application of AI are required; thus, creating AI service centers was the logical step for making AI in Germany a joint effort.
A Session on “AI Applications in Clinical Practice, Challenges, and Successes” was organized by Prof. Dr. Anne-Christin Hauschild (Head of scientific research group Clinical Decision Support), Dr. Nicolai Spicher (Head of scientific research group Biosignal Processing), and Dr. Zully Ritter (scientific research group Clinical Decision Support). The session started with a keynote by Michael Dietrich, a senior developer of AI solutions from DFKI Berlin, in which the chances and risks of AI in medicine were presented, using examples of current and recently developed solutions. Additionally, four talks involving predictive models for heart failure using wearables (Hempel P. et al.), a contribution of our institute, as well as the usage and benchmarking of newly developed methods like forward-forward propagation (Scodellaro R. et al.), patient-specific support AI mental models using the example of knee rehabilitation (Janzen S. et al.) and how to mitigate privacy issues when AI models are implemented in clinical set-ups (Schrod S. et al.). It was clear that beyond the challenges of implementing AI solutions, those well-known and new ones, data sharing, privacy security, and patient acceptance concerns need to be addressed during the process. Examples from clinical practice evidence that testing covers the technical aspects, and those involving the different users, including the patient's acceptance, are fundamental.
A workshop on “Explainable AI in Clinical Decision Support Systems” was organized by Prof. Dr. Anne-Christin Hauschild (Head of scientific research group Clinical Decision Support) and Dr. Zully Ritter (scientific research group Clinical Decision Support) in collaboration with Miriam Maurer (scientific research group Clinical Decision Support). Participants from the academy, industry, and students were presented. The workshop consisted of a theory and a hands-on part. The theory part covered an introduction to XAI, its methods, taxonomy, recommended XAI post-hoc methods according to the research question (collected or available data), and current commercial XAI solutions for machine learning and deep learning. Two hands-on exercises included understanding and applying XAI to tabular clinical data (heart failure) and time series data (ECG). A discussion concerning recommendations on using it in the praxis closed the workshop. It is already widely known that XAI is required to increase the trust of AI solutions in healthcare, especially in healthcare staff and patients. XAI allows understanding (visualizations, quantifications) of how and why the model has selected specific data to make a prediction. However, considering that the AI-Act explicitly includes XAI when developing and commercializing AI solutions, the necessity of offering such workshops, including hands-on exercises, is a call for joining efforts to make it affordable for all.
