Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.37213

Retrospective macromolecular pool fraction mapping from conventional clinical MRI: a physics-guided self-supervised deep learning framework

Jelmer van Lune·Miha Fuderer·Oscar van der Heide·Matteo Maspero·Martin B. Schilder·Jan Willem Dankbaar·Stefano Mandija·Alessandro Sbrizzi
View original source ↗Read source document ↗Export metadata ↓

Abstract

Macromolecular pool fraction (MPF) is a quantitative imaging biomarker sensitive to myelin integrity and other changes in tissue macromolecular content, but specialized acquisitions and reconstruction methods limit its use in large-scale studies. We present a physics-guided, self-supervised deep learning framework for retrospectively estimating MPF directly from conventional clinical brain MRI ($T_1$-weighted, $T_2$-weighted, and fluid-attenuated inversion recovery [FLAIR]) without paired quantitative reference scans. The framework embeds an analytical two-pool magnetization transfer signal model that related sequence-averaged radiofrequency transmit power to apparent longitudinal relaxation. To accommodate heterogeneous clinical protocols, the network conditions intermediate features on acquisition parameters via adaptive instance normalization. It was trained and evaluated on a multisite, multivendor, multiprotocol dataset of 7,705 MRI sessions (3 T). Across 1,237 test sessions, mean MPF was $0.182 \pm 0.007$ in white matter and $0.144 \pm 0.009$ in gray matter, with limited variation across vendor-by-protocol groups (between-group coefficient of variation: 2.65% and 5.63%, respectively). In 70 sessions with demyelinating-disease labels, MPF was 35.6% lower in segmented lesions than in normal-appearing white matter. MPF maps from two different clinical protocols acquired in a healthy volunteer showed a mean voxel-wise absolute percentage difference of 7.7% and a concordance correlation coefficient of $0.888$, indicating good cross-protocol consistency. These results suggest that magnetization transfer-informed self-supervised learning enables stable MPF mapping from conventional clinical MRI, providing a scalable approach to investigate demyelinating pathology in both existing imaging archives and prospectively acquired conventional clinical scans without dedicated qMT acquisitions.

Explore related subjects

physics.med-phcs.CVcs.LG

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jelmer van Lune, Miha Fuderer, Oscar van der Heide, Matteo Maspero, Martin B. Schilder, Jan Willem Dankbaar, Stefano Mandija, Alessandro Sbrizzi. 2026-10-27. Retrospective macromolecular pool fraction mapping from conventional clinical MRI: a physics-guided self-supervised deep learning framework. https://arxiv.org/abs/2610.37213

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Can future reirradiation justify upfront proton therapy? An extension of model-based proton therapy selection beyond primary-course NTCP benefit

Objective: Current model-based proton therapy selection values normal tissue sparing only through its immediate reduction in normal tissue complication probability (NTCP). If reirradiation later adds further dose to the normal tissue, sparing achieved during primary treatment may become more clinically valuable. We investigated whether, and under which conditions, this contingent benefit could justify upfront proton therapy. Approach: We formulated a reirradiation-aware criterion that weights the NTCP benefits of upfront proton therapy after primary treatment alone or followed by reirradiation by their respective probabilities. Proton therapy is recommended when the expected benefit reaches a referral threshold T. Assuming a common logistic NTCP model for both scenarios, we derived screening conditions from the paired primary photon and proton plans to identify patients for whom future reirradiation could justify additional referral. We then quantified the minimum reirradiation probability and admissible added risk-score contributions for additional referral, and mapped referral patterns across primary profiles and reirradiation scenarios. Main results: Additional referral occurred for patients with substantial primary normal tissue sparing but a small immediate NTCP reduction because both plans lay on a flat region of the NTCP curve. In these cases, reirradiation could add sufficient normal tissue dose for the retained sparing to yield a larger NTCP benefit. Higher reirradiation probability expanded eligibility, whereas greater exposure could reduce the NTCP benefit as both predictions approached saturation. Significance: The proposed framework identifies well-defined patient profiles for which upfront proton therapy may be justified even if the NTCP benefit of proton therapy in the first treatment is below the threshold T.

physics.med-ph↗
arXiv2026

Cerebral Artery Wall Displacement and Strain from 4D MRI

Arterial stiffness is an established marker of vascular ageing and cardiovascular risk, yet the local, patient-specific mechanics of the cerebral arteries remain difficult to measure in vivo. Wall deformation over the cardiac cycle is sub-millimetre, rapid and spatially heterogeneous, and is superimposed on larger rigid-body motion. This work adapts a registration-based kinematic framework, originally developed for abdominal aortic aneurysms, to cerebral arteries imaged with 7 T 4D phase-contrast MRI. Deformable image registration between diastole and systole was used to measure wall-normal displacement over the cerebral artery network, primarily the Circle of Willis and connected communicating arteries. Following this, transverse centreline motion was removed to separate local wall deformation from rigid-body vessel motion before estimating circumferential strain. Across 17 datasets, the 99th percentile wall-normal displacement was 0.256 +/- 0.037 mm, 35 % of the acquired voxel dimension, and the 99th percentile circumferential strain was 8.67 +/- 2.40 %. Correction of circumferential strain to exclude rigid translation motion reduced the 99th percentile strain estimation in the example case from 28.1 % to 10.6 %, as a significant portion of the apparent wall displacement corresponds to bulk motion of the vessel. The pipeline produces continuous displacement and strain maps of the intracranial arterial network, although the values are sensitive to segmentation accuracy, resampling and registration parameterisation and require future ground-truth verification.

physics.med-ph↗
arXiv2026

Just-in-time volumetric imaging: Learning 2D-3D thoracic motion models during image-guided interventions

Despite two decades of innovation, cutting-edge bronchoscopic techniques still fail to yield clear diagnostic decisions in 30% of cases and miss malignancies in 20% of cases. Intraprocedural 3D imaging drastically improves both tool positioning and diagnostic yield. However, each 3D spin represents a static snapshot of an anatomy that is continuously evolving as the patient is ventilated throughout the procedure. Here, we propose just-in-time volumetric imaging, in which a patient-specific 2D--3D motion model is trained on pre-intervention scans before it is needed and then supplies volumetric images from single fluoroscopic projections during sampling. Given the inspiration-hold CBCT already acquired for navigation and an additional motion-inclusive CBCT, our optimization regime named VoxelMap-DualScan (VM-DS), solves a low-rank motion model by alternating least squares and then amortizes it into a projection encoder trained on synthesized projections with known motion, all within a 10-minute window. On eight held-out cases from the public DIR-Lab 4DCT dataset, VM-DS achieved a mean landmark target registration error of 2.8 mm, a structural similarity index of 0.95 and a peak signal-to-noise ratio of 37.6 dB. Accuracy was maintained when reducing the number of projections used for training from 227 to 57. Inference takes 1.5 ms per projection, making the framework suitable for real-time use during image-guided interventions. Code will be made available at https://github.com/Image-X-Institute/Voxelmap/tree/VM-DS

physics.med-ph↗