Jorge Reis-Filho: A New Framework for Evaluating Medical Intelligence
Jorge Reis-Filho And Jakob Nikolas Kather

Jorge Reis-Filho: A New Framework for Evaluating Medical Intelligence

Jorge Reis-Filho, Chief of AI for Science Innovation, Enterprise AI Unit at AstraZeneca, shared a post by Jakob Nikolas Kather, Professor of Clinical Artificial Intelligence at Dresden University of Technology, on LinkedIn, adding:

”Many congratulations to Jakob Nikolas Kather, Dyke Ferber and the whole team for the Nature paper on MIRA out today.

This is a remarkable step for agentic AI in clinical practice.

The agent itself will deservedly get plenty of attention. For me, the most important contribution is the framework of task-specific and end-to-end benchmarks for the full clinical decision pathway.

Clinical medicine is a sequence of decisions under varying levels of uncertainty. Gathering evidence, interpreting it, acting on it and revising as new information emerges. The ‘correct diagnosis’ or ‘right treatment’ is one point along a longer trajectory. MIRA lets us benchmark the process itself. That is where medical intelligence actually lives.

Kudos to the team for addressing a crucial bottleneck in moving these systems forward responsibly.

It should come as no surprise that this paper appears back-to-back in Nature with a convergent study from Google and Google DeepMind.

When an academic group and a hyperscaler reach similar conclusions independently, it is a clear signal, undoubtedly. The frontier is about framing, scaffolding, tools and rigorous end-to-end evaluation and benchmarking.

This is commendable progress. Prospective clinical validation is the next logical step. As we start thinking about real-world deployment, hospital infrastructure and reliable access to the underlying models will be as important as the models themselves.

Kudos again to Jakob Nikolas Kather, Dyke Ferber and the entire team for pushing the boundaries of what we can do.

What will it take for hospitals to benchmark pathway-level medical intelligence prospectively, safely and at scale?”

Quoting Jakob Nikolas Kather’s post:

”Our paper ‘Towards autonomous medical artificial intelligence agents’ is out in Nature today! It was led by Dyke Ferber who did the real work here.

We present MIRA, an autonomous AI agent that runs end to end through a clinical patient case. We build it on top of LLMs and provide it with scaffolding, tools, and most importantly, a thorough evaluation against human doctors.

We show that our agent can solve a clinical case in a multistep process. It takes the history, orders labs and scans, selects medications, decides on procedures, and triages for admission.

For me the most important part is that we can now benchmark humans against AI for the whole decision pathway, not just the end result. Our evaluation shows very good results, almost no severe errors, these AI agents perform just really well on difficult medical cases.

As we move now to implement this in real clinical routine, and run clinical trials about it, we must first fix the infrastructure in our hospitals (and also ensure we still have access to the underlying LLMs…).

Kudos also to our colleagues from Google and Deepmind who published a very similar study back-to-back with us!

We are happy that with <<0.01% of Google’s RandD budget we can contribute to this space together.

Thanks to all co-authors, our institutions and also the great Eric Topol for covering us in his newsletter. Congratulations to Dyke and the whole team!”

Title: Towards Conversational AI for Disease Management

Authors: Valentin Liévin, Anil Palepu, Wei-Hung Weng, Khaled Saab, David Stutz, Yong Cheng, Kavita Kulkarni, S. Sara Mahdavi, Joëlle Barral, Dale R. Webster, Katherine Chou, Avinatan Hassidim, Yossi Matias, James Manyika, Ryutaro Tanno, Vivek Natarajan, Adam Rodman, Tao Tu, Alan Karthikesalingam, Mike Schaekermann

Read The Full Article.

Jorge Reis-Filho: A New Framework for Evaluating Medical Intelligence

Title: Towards autonomous medical artificial intelligence agents

Authors: Dyke Ferber, Lars Hilgers, Christiane Höper, Benedict Kinny-Köster, Jan-Niklas Eckardt, Katharina Egger-Heidrich, Marius Bill, Martin M. K. Schneider, Jan Clusmann, Lejla Kadric, Marcel Oehme, Maximilian Mayrhofer-Schmid, Alexander Oeser, Georg Wölflein, Isabella C. Wiest, Jan Moritz Middeke, A. John Iafrate, Daniel Truhn, Dirk Jäger, Jakob Nikolas Kather

Read The Full Article.

Other Articles Featuring Jorge Reis-Filho And Jakob Nikolas Kather on OncoDaily.