Heidelberg, Tübingen, Erlangen, MR physicist, science communicator, CEST and Bloch-McConnell fan, Pulseq fan, MR-zero Prof @FAU_Germany
Joined March 2015
- Tweets914
- Following378
- Followers526
- Likes2.1K
Pinned Tweet
Did you hear about autoresearch by @karpathy?
tinyurl.com/385axffj
We recreated it, but for MRI.
We let AI agents autonomously write, simulate, and refine MRI sequences -- then compete on an MR autoresearch leaderboard.
Watch agents improve MRI (FLAIR-SE-EPI): 1/5
Moritz Zaiss retweeted
Pay-for-a-Paper? How Fastcode AI Packs a $100 K Contract into a Conference Publication
A small extension of Lars Hansons beautiful web Bloch Simulator allows you now to load Pulseq files.
Now you can watch the spins dancing to an MRI sequence, here for example a spiral TSE. Work in progress mzaiss.github.io/Bloch-Simul…
Original: drcmr.dk/BlochSimulator/
Moritz Zaiss retweeted
Grigori Perelman was both the first and the last human to solve a Millennium prize problem.
Moritz Zaiss retweeted
For years I have been documenting the depiction of magnetic resonance imaging (MRI) in pop culture. The technology seems to fascinate the public, but it is often portrayed with errors, particularly when it comes to the underlying physics. From computed tomography (CAT) scans being confused for MR images, to metallic objects flying across hospital hallways, entertainment media routinely override real physics for dramatic effect. So I thought to myself, what if I could turn pop culture’s fascination with MRI into a teaching moment for my MRI class this year? 1/n
Moritz Zaiss retweeted
Watch my new sci-fi film SPECTRUM.
The entire project took three weeks to finish and was made by one person, with the help of various AI tools used throughout the production. I’ve listed all the tools at the end of the film.
Plot: In the midst of World War II, soldiers encounter something that should not exist: a man in color, in a world that has only ever known black and white.
Written and directed by Hashem Al-Ghaili.
Moritz Zaiss retweeted
Foundation Models for Pan-cancer Segmentation in CT Images (Codabench Challenge #7149)
Moritz Zaiss retweeted
XAI has a dichotomy: real systems are too complex to fully specify, toy systems are too simple to matter.
Together with Siming Bayer and Patrick Krauss, I just put a paper on arXiv that breaks it open by making the Atari 2600 VCS itself differentiable.
Moritz Zaiss retweeted
When the AI Writes, the Detectors Sweat: Three Hours of Agentic Research in the AI-Text Arms Race
akmaier.substack.com/p/when-…
Moritz Zaiss retweeted
Samira's work made the cover of @NatureNeuro this month 🎉
nature.com/neuro/volumes/29/…
I also want to highlight two recent comments that engage with our quantitative approach to interpreting the BOLD signal; both are worth reading (see below).
Moritz Zaiss retweeted
When AI Starts to Write Code That Actually Works: LLMs Master the Art of Medical Imaging
akmaier.substack.com/p/when-…
Moritz Zaiss retweeted
Medical Valley and BAIOSPHERE Medical Join Forces for a Special Anniversary Celebration
Paper submission deadline: 1st of May!
We are delighted to announce that BAIOSPHERE Medical 2026 will collaborate with Medical Valley EMN e. V. as part of Medical Valley’s 20th anniversary
Moritz Zaiss retweeted
BAIOSPHERE Medical 2026 announces its venue: the new FAU Conference Center
Abstract Submission Deadline: May 1st!
We are delighted to announce that BAIOSPHERE Medical 2026 will take place at the new FAU Conference Center, Freyeslebenstr. 1, Erlangen, on June 9th, 2026.
Moritz Zaiss retweeted
BAIOSPHERE MEDICAL 2026 Keynote: Mengyun Qiao, UCL
Deadline May 1st!
Beyond Digital Twins: Generative and Agentic Cardiac Modelling for Clinical Discovery
Moritz Zaiss retweeted
BAIOSPHERE MEDICAL 2026: Registration Now Open – Free Registration for Accepted Abstracts and Highlights
BaioMed 2026 will take place in Erlangen, Germany, on June 8–9, 2026, bringing together clinicians, imaging scientists, AI researchers, computer scientists, engineers,
We ran the MR autoresearch challenge across 3 model generations. Agents discovered strategies on their own: multi-shot EPI, adaptive recon, differentiable optimization of sequence parameters.
Newer models = better MR agents. And newest models beat our human expert solution. 4/5
The concept: give agents MR domain tools + a quality metric, let it iterate. Research inherently relies on validation and testing -- that's the loop agents are good at.
Built on Pulseq/PyPulseq/MR-zero 5/5
Blog: mr-physik.med.fau.de/2026/04…
Paper: arxiv.org/abs/2604.13282
@maier_ak
