MLOps Community Healthtech by Delphyr @ Philips
Join us for a special healthtech edition of the MLOps Community Amsterdam in partnership with Delphyr and Philips, where we dive into the world of AI for healthcare, exploring how ML models and LLMs are being applied to solve real-world challenges and transform patient outcomes.
📅 Agenda
17:30 - 18:15: Walk-in 🚶♂️ + Dinner
18:15 - 18:30 Opening words
18:30 - 19:00: Riaan Zoetmulder
19:00 - 19:30: Tim de Boer
19:30 - 19:45: small break.
19:45 - 20:15: Monu Krishnan
20:15 - 21:00: Networking + drinks 🍹
Talks:
Artificial Intelligence for Image Quality Improvement: Challenges & Perspectives - Riaan Zoetmulder (Philips)
AI in medical imaging is often discussed in terms of supervised tasks like segmentation, classification, or detection. But improving image quality is a different kind of problem. In image-guided therapy, image quality is central to safe and effective clinical use, yet it is only partially captured by physical metrics such as noise, radiation dose, or contrast. It also includes qualitative perceptual aspects that are harder to measure and may vary between individuals.
In this talk, I will use image-guided therapy systems as a case study to show how image quality has been improved and validated before the adoption of deep learning. Next, I will discuss what changes as these systems become more AI-driven. The interesting part is not that we already have all the answers, but that incorporating AI challenges the assumptions underlying the validation and post-market surveillance.
This then raises a number of open questions: how do we evaluate image quality when part of the target is subjective, how do we deal with domain shift across patients and acquisition settings, and how do we ensure safe deployment in a regulated environment? Finally, how do we make sure the physician trusts our algorithms?
In our setting, the hardest problems and trade-offs do not arise when developing the model, but when trying to validate and deploy it to over 12500 machines deployed in a time-critical environment.
2. Building Generative AI for Healthcare - Tim de Boer (Delphyr)
Building generative AI for healthcare presents challenges not usually encountered elsewhere. EHRs are complex systems carrying decades of clinical workflows, so integrating with them takes partnership and effort from both sides.
Integrations that last are built as long-term collaborations with the teams behind these systems, not as quick workarounds that fail the next time something upstream changes. The data reflects how care is actually delivered: it is narrative, contextual, and full of abbreviations that might vary between specialties or departments. The stakes are high: if an LLM confidently hallucinates the wrong dosage, the consequences are severe. Regulations like the MDR exist to protect patients and shape almost every decision you make as engineer.
In this talk I’ll walk through 10 things to consider when building generative AI for healthcare.
3. - Monu Krishnan (Skinvision)
TBA
Promo code:
Use the following promo code to get a free subscription to Skinvision: SV26-MLOPS