Medicine Henry Quentir Medicine Henry Quentir

Tohoku University Researchers Use Quantum Annealing to Choose Which MRI Lines a Knee Scan Acquires: What arXiv 2611.25625 Tested on fastMRI Images, 19 November 2026

How Huang and Ohzeki turned the choice of MRI k-space lines into a QUBO problem

Quentir Medicine Monitor

Evidence-based insights for quantum medicine.

A preprint that Xiaoxin Huang and Masayuki Ohzeki of Tohoku University posted on 19 November 2026 treats a routine decision in accelerated MRI as an optimization problem for an annealer: which lines of raw scanner data a shorter scan should acquire and which it may skip. The authors scored every candidate line by how much information it carries about image quality, combined those scores into a single binary optimization model, and solved it twice, once with classical simulated annealing and once with D-Wave's quantum-classical hybrid solver. On knee images from the public fastMRI collection, the resulting sampling masks reconstructed images with lower average error than four standard sampling patterns at every scan budget the authors tested.

The clinical motive is scan time. A full MRI acquisition can take long enough to be difficult for people with claustrophobia and for patients who cannot lie still, and the authors open their paper with exactly that burden. Acquiring fewer data lines shortens the scan, but every skipped line removes information the reconstruction software must do without. The choice of lines therefore decides how much image quality a department gives up for each minute it saves. The paper proposes a new way to choose the lines and tests it in simulation; no real examination has yet been shortened with it.

An MRI scanner acquires its signal in k-space, the frequency representation of the image. Lines near the center of k-space carry overall contrast and shape; lines toward the edge carry edges and fine detail. Many accelerated MRI methods use fixed sampling rules. One widely used example is the pseudorandom variable-density sampling that Michael Lustig, David Donoho and John Pauly proposed in their 2007 compressed sensing paper in Magnetic Resonance in Medicine, which samples the center densely and the periphery sparsely. Such rules are written in advance and do not adapt to images of the anatomy being scanned. Later studies, which the preprint reviews, made the mask itself an optimization target: Gözcü and colleagues searched for masks by their average performance on training images, and the LOUPE method learns a sampling distribution jointly with a reconstruction network.

Read More