Stephan Schug

Healthcare is risking high penalties. If they continue to ignore the EU AI Act. Here are the key aspects of the AI Act for healthcare to consider: 1. Risk-Based Approach Categorizes AI systems based on their potential risk level, from minimal to prohibited. Appropriate regulation based on the potential impact of AI systems, with stricter requirements for higher-risk applications. Failing to properly categorize and comply with risk-based requirements could lead to legal issues and potentially unsafe AI use in healthcare. 2. High-Risk AI Systems Many healthcare AI applications are classified as high-risk, thus strict compliance requirements. It ensures safety and ethical standards for AI systems that could significantly impact patient health and safety. Non-compliance could result in severe penalties and reputational damage. 3. Transparency Requirements Mandates clear communication about AI systems' capabilities, limitations, and decision-making processes. It promotes trust and understanding between healthcare providers, patients, and AI systems. Lack of transparency could lead to misuse of AI systems and ruin patient trust. 4. Data Governance and Quality Strict standards for data management, including quality, integrity, and security. Ensures AI systems are trained on reliable data, crucial for accurate healthcare outcomes. Poor data governance could lead to biased or inaccurate AI outputs, potentially harming patients. 5. Human Oversight Requirement for human supervision and intervention in decision-making processes. Healthcare professionals have critical role in patient care and decision-making. Over-reliance on AI without proper human oversight could lead to errors and ethical issues. 6. Conformity Assessments Third-party evaluations required for high-risk AI systems. Ensures AI systems meet safety and performance standards before deployment. No proper assessments could prevent market access and lead to legal issues. 7. Ethical Considerations Focus on respecting human dignity, autonomy, and privacy in AI development and use. Ensures AI in healthcare aligns with fundamental rights and ethical principles. Violations could lead to legal consequences and loss of public trust. 8. Innovation Support Services like regulatory sandboxes allow real-world testing of AI systems. Supports continued innovation in medical AI while maintaining safety standards. Not using these tools could hinder the development of beneficial AI tech. 9. AI Literacy Require providers and deployers to ensure staff know responsible AI use. Providing individuals with the knowledge and skills needed to understand, use, and critically evaluate AI technologies. AI literacy helps organizations meet the AI Act's requirements for staff competence in AI systems. If the Act is not taken seriously, it can lead to penalties.  Up to €35 million or 7% of global turnover. So time to start thinking about a strategy. What are you doing to prepare?

369

35 Kommentare

📢 Attention Digital Health Experts: A Key Study on AI in Healthcare Dear digital health professionals, if you haven't had the chance to explore it yet, our insightful study on the potential of Artificial Intelligence in healthcare is a must-read. With European healthcare systems facing significant challenges, AI could be a transformative solution. Our study, released in July, highlights: 🔍 Key Barriers to Adoption: Addressing data limitations, and tackling legal, organizational, and cultural challenges. 🏥 Real-World Success Stories: Learn from hospitals successfully integrating AI technologies. 🛠️ A Proposed Framework: Detailed strategies to monitor and accelerate AI integration across EU healthcare systems. This is an invitation to dive into how we can safely, ethically, and effectively lead AI innovation while protecting patients' rights. Don't miss out on driving the future of digital health—check out the study if you haven't yet: 👉 https://lnkd.in/eN8uNEei #HealthUnion #DigitalHealth #EHDS #AI

416

9 Kommentare

🧠 𝗔𝗻 𝗶𝗻𝘁𝗲𝗿𝗲𝘀𝘁𝗶𝗻𝗴 𝗿𝗲𝗮𝗱 (𝘄𝗶𝘁𝗵 𝗮 𝘁𝘄𝗶𝘀𝘁 𝗮𝘁 𝘁𝗵𝗲 𝗲𝗻𝗱)! Check out this new publication in JAMA Network Open about AI-generated hospital course and discharge summaries. Key quotes and my 2 cents are below: 𝗜𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝗰𝗲 “Hospital course (HC) summarization represents an increasingly onerous discharge summary component for physicians. Literature supports large language models (LLMs) for HC summarization, but whether physicians can effectively partner with electronic health record–embedded LLMs to draft HCs is unknown.” 𝗢𝗯𝗷𝗲𝗰𝘁𝗶𝘃𝗲𝘀  “To compare the editing effort required by time-constrained resident physicians to improve LLM- vs physician-generated HCs toward a novel 4Cs (𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦, 𝘤𝘰𝘯𝘤𝘪𝘴𝘦, 𝘤𝘰𝘩𝘦𝘴𝘪𝘷𝘦, 𝘢𝘯𝘥 𝘤𝘰𝘯𝘧𝘢𝘣𝘶𝘭𝘢𝘵𝘪𝘰𝘯-𝘧𝘳𝘦𝘦) HC.” 𝗗𝗲𝘀𝗶𝗴𝗻, 𝗦𝗲𝘁𝘁𝗶𝗻𝗴, 𝗮𝗻𝗱 𝗣𝗮𝗿𝘁𝗶𝗰𝗶𝗽𝗮𝗻𝘁𝘀  “Quality improvement study using a convenience sample of 10 internal medicine resident editors, 8 hospitalist evaluators, and randomly selected general medicine admissions in December 2023 lasting 4 to 8 days at New York University Langone Health.” 𝗘𝘅𝗽𝗼𝘀𝘂𝗿𝗲𝘀    “Residents and hospitalists reviewed randomly assigned patient medical records for 10 minutes. Residents blinded to author type who edited each HC pair (physician and LLM) for quality in 3 minutes, followed by comparative ratings by attending hospitalists.” 𝗠𝗮𝗶𝗻 𝗢𝘂𝘁𝗰𝗼𝗺𝗲𝘀 𝗮𝗻𝗱 𝗠𝗲𝗮𝘀𝘂𝗿𝗲𝘀  “Editing effort was quantified by analyzing the edits that occurred on the HC pairs after controlling for length (percentage edited) and the degree to which the original HCs’ meaning was altered (semantic change). Hospitalists compared edited HC pairs with A/B testing on the 4Cs (5-point Likert scales converted to 10-point bidirectional scales).” 𝗖𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗥𝗲𝗹𝗲𝘃𝗮𝗻𝗰𝗲    “Electronic health record–embedded LLM HCs required 𝗹𝗲𝘀𝘀 editing than physician-generated HCs to approach a quality standard, resulting in HCs that were 𝗰𝗼𝗺𝗽𝗮𝗿𝗮𝗯𝗹𝘆 𝗼𝗿 𝗺𝗼𝗿𝗲 complete, concise, and cohesive, but contained 𝗺𝗼𝗿𝗲 confabulations … this study supports the feasibility of a physician-LLM partnership for writing HCs and provides a basis for monitoring LLM HCs in clinical practice.” 𝗠𝘆 𝟮 𝗖𝗲𝗻𝘁𝘀 AI can significantly boost our abilities but only with human experts in the driver’s seat. The moment we hand over the wheel to the AI is the moment it drives us into a wall. As the authors note: “While these tools have exciting potential, their integration into clinical practice requires EHR vendors and health systems to support physicians with education, local validation, and safety guardrails.” ____________________________ (Author credits: William Small, MD, MBA et al.; source in the comments).

31

7 Kommentare

🚨 Delighted to share our latest publication in Mayo Clinic Proceedings: Digital Health on “Deep Learning Applications in Clinical Cancer Detection: A Review of Implementation Challenges and Solutions” https://lnkd.in/gj2jPZtq In this comprehensive review, my co-authors and I analyzed over 1,400 studies from 2018–2024, spanning imaging, genomics, pathology, and multimodal diagnostics. We highlight: ✅ The tremendous promise of deep learning (DL) in improving early cancer detection ✅ Persistent challenges in clinical adoption—including interpretability, bias, and regulatory hurdles ✅ Cutting-edge solutions, such as federated learning, explainable AI, synthetic data generation, and real-world trials ✅ A call for interdisciplinary collaboration to ensure AI tools are ethically, equitably, and effectively deployed in diverse healthcare settings I have always supported the integration of AI into oncology especially when I served as MD/CEO of the National Cancer Centre Singapore from 2017-2023 and even took a course with the MIT Sloan School of Management on of AI in Healthcare back in 2020. So when I had two junior college students Isaiah Yao and Dong Min approach me for an elective posting I set them a task with some directions on getting this paper done. They surprised me with excellent work done on this and this paper is a culmination of many months of work together on this. Deep learning is not a distant promise—it’s already transforming diagnostics. The question now is: How do we scale it safely and fairly for all? Grateful to my co-authors and to REMEDIS for supporting this work. Also to Yinxia Chao of Duke-NUS Medical School for referring the students to me. #AIinHealthcare #CancerDetection #DeepLearning #MayoClinic #DigitalHealth #OncologyInnovation #ExplainableAI #NCCS #REMEDIS #artificialintelligence Mayo Clinic SingHealth

82

6 Kommentare