Foundation training for employees with and without prior knowledge: compact, practice-oriented and framed by current regulation - from AI/ML basic terms to bias & fairness to the EU AI Act and practical prompting.
The foundation training is aimed at employees with and without prior knowledge and provides central training content to understand, assess and use AI in the healthcare sector responsibly. It was created and editorially checked on the basis of peer-reviewed sources and applicable EU regulations.
The material combines technical principles (AI/ML/Deep Learning, algorithms, foundation models) with clinical practice (fields of application, metrics, prompting), ethical principles and legal framework (EU AI Act, GDPR, MDR). It is important to understand that AI is not a neutral tool – every system reflects the data and decisions from which it was built.
Differentiation of terms, ML learning types (supervised/unsupervised/reinforcement), classical algorithms (decision trees to neural networks), foundation models and LLMs. Strengths and limitations in the clinical context.
Radiology and imaging, clinical decision support, early detection, automatic coding, monitoring, drug development and genomics. Opportunities must always be assessed against error risks, bias, validation limits and the specific clinical use context.[3,7,9]
Trade-off between sensitivity and specificity, the importance of prevalence (base-rate problem), calibration and why models can perform significantly worse outside their training and validation context.[7,10]
Representation bias, historical bias, measurement bias, label bias and subgroup performance. XAI methods such as SHAP or Grad-CAM can support analysis and communication, but they do not replace validation, fairness testing and clinical assessment.[7,8]
7 EU HLEG requirements for trustworthy AI,[1] the EU AI Act with its risk-based approach, prohibited practices, high-risk obligations and transparency requirements,[4,5] GDPR for personal health data and MDR/IVDR when there is an intended medical purpose.[3,4,6] In addition, the European Health Data Space (EHDS, Regulation (EU) 2025/327) addresses the use of health data.
Effective prompting (context, role, requesting sources, addressing uncertainty), critical review of generative outputs, transparency when using AI and separate documentation of AI recommendations and human decisions.[2,4,9]
The material addresses a practical gap in many healthcare organizations: employees who work with AI outputs or accompany AI implementation projects need basic knowledge of data quality, validation, bias, human oversight and regulatory classification. Without this knowledge, the strengths and limitations of real AI systems are difficult to assess, with consequences for patient safety, compliance and trust.[1,7,9]
The content follows the European regulatory framework and is prepared for a non-technical audience. Statements with technical or regulatory weight are supported by the sources listed below.
Unless otherwise stated, retrieved: April 2026.
This material was created with the greatest possible care. It is not a substitute for legal, medical or professional advice. No guarantee for completeness or topicality.
Parts of this document were created and editorially checked with the support of generative AI. According to EU AI Act Art. 50, the use of AI is transparently pointed out.