Synthesizing Post-Acetazolamide Cerebral Blood Flow Maps from Baseline MRI in Moyamoya Using 3D Generative AI

Julia Huang; Camila Gonzalez; Rydham Goyal; Aja Zou; Sasha Alexander; Michael Moseley; Moss Y. Zhao; Gary K. Steinberg

eess.IVAug 14, 2026

Synthesizing Post-Acetazolamide Cerebral Blood Flow Maps from Baseline MRI in Moyamoya Using 3D Generative AI

Authors: Julia Huang, Camila Gonzalez, Rydham Goyal, Aja Zou, Sasha Alexander, Michael Moseley, Moss Y. Zhao, Gary K. Steinberg

Organizations: Department of Computer Science, Stanford School of Engineering, Stanford, CA 94305 · Department of Radiology, Stanford School of Medicine, Stanford, CA 94305 · Department of Anesthesia, Intensive Care Medicine, and Pain Medicine, Medical University of Vienna, Vienna, Austria 1090 · Department of Neurosurgery, Stanford School of Medicine, Stanford, CA 94305

Abstract

For patients with Moyamoya disease, impaired cerebrovascular reserve (CVR) is an important hemodynamic criterion for recommending extracranial-to-intracranial bypass surgery. Standard CVR assessment in this cohort uses paired arterial spin labeling (ASL) perfusion MRI acquired before and after acetazolamide (ACZ). When ACZ is contraindicated or avoided, the post-ACZ cerebral blood flow (CBF) map needed for hemodynamic assessment is unavailable. We propose CAE3D, a deterministic 3D conditional autoencoder that synthesizes post-ACZ CBF maps directly from pre-ACZ ASL input. We evaluated CAE3D against ten comparators, including deterministic and diffusion-style 3D baselines, a 2D contextual baseline, and frozen-encoder foundation-model adapters. CAE3D achieved the lowest held-out MAE (0.066), with SSIM 0.80 and PSNR 24.0 dB, and near-zero full-brain mean bias. Its MAE advantage was statistically significant over seven of eight trained-from-scratch baselines, excluding the 2D CAE_2D comparator; its SSIM and PSNR advantages were significant over all eight. Regional delta-CBF predictions compressed the dynamic range in high-response territories. These results establish the retrospective feasibility of post-ACZ CBF synthesis in patients who completed the standard two-scan protocol. Extension to ACZ-contraindicated patients, who were not represented in this cohort, requires external and prospective validation.

Figures & tables

Panel of Figure 1 — Figure 1: Established clinical pathway. The protocol used in this cohort requires two ASL scans (pre- and post-ACZ) bracketing an intravenous acetazolamide administration.

Panel of Figure 2 — Figure 2: Retrospective synthesis task evaluated in this study. CAE3D predicts an independently normalized post-ACZ map from the pre-ACZ input. This diagram illustrates the potential future workflow motivation, not a clinically validated replacement for ACZ imaging.

Panel of Figure 3 — Figure 3: CAE3D architecture. Skip-connected 3D encoder-decoder (three downsampling levels).

Figure 4: Image context. Pre-ACZ, post-ACZ GT, and post-minus-pre change (middle axial slice, one subject). Image panels use intensity windowing for visualization ( $v_{\max}$ set to the 99th percentile of non-zero voxels); see the plotting scripts in the supplementary material for exact details.

Figure 5: Vascular territory atlas ( Liu et al., 2023 ) on a representative pre-ACZ slice; defines regional summaries (§ 3.4 ). Windowed for visualization as in Figure 4 .

Figure 6: Qualitative reconstruction (CAE3D, held-out test set): best, median, worst PSNR subjects (top to bottom). Columns: pre-ACZ, prediction, GT, error (middle axial slice). Per-subject MAE/SSIM/PSNR in row labels.

Figure 7: Guided backpropagation (CAE3D, one test subject): pre-ACZ, attribution overlay (brighter = stronger influence), GT post-ACZ. Full guided-backpropagation maps are provided in the supplementary material.

Model	Primary seed	Checkpoint-selection rule
CAE3D	42 (predesignated)	Highest validation PSNR, no early stopping
CAE3D-ES	42 (predesignated)	Highest validation PSNR, early stopping (10-epoch plateau)
ResNet_3D	42 (predesignated)	Highest validation PSNR
CAE_2D	42 (predesignated)	Highest validation PSNR
FNO_3D	42 (predesignated)	Highest validation PSNR
Patch_3D	42 (predesignated)	Highest validation PSNR

Table 5: Primary-checkpoint selection rule for each in-house model’s rows in Tables 1 and 6 . Seed 42 was predesignated for every model before training (not chosen post hoc from among the three seeds); within that seed, the checkpoint with the highest validation PSNR was retained, with no test-set involvement in selection. Exact checkpoint filenames and configurations are documented with the code release.

Medical Image Synthesis

Medical image synthesis creates, translates, reconstructs, and edits clinical images to fill missing data and support analysis.

See all 310 papers

One Frame, Full Heartbeat: ECG-Free 4D Cardiac Cine MRI Synthesis via Radial-Decomposed Flow Matching
Oct 6, 2026Shiyi Wang, Ruochen Sun, Xiang Li +2Medical Image Synthesis Cardiac MRI
Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA · Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA · Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA
ICHOR: A Robust Representation Learning Approach for ASL CBF Maps with Self-Supervised Masked Autoencoders
Mar 5, 2026Xavier Beltran-Urbano, Yiran Li, Xinglin Zeng +10Medical Image Representation Learning Self-Supervised Pre-Training
University of Pennsylvania, Philadelphia, USA · University of Maryland School of Medicine, Baltimore, USA
One Evaluation, Any Operating Point: Hypernetwork-Amortized MeanFlow for 3D MRI Reconstruction
Sep 15, 2026Ruibo WangAccelerated MRI Reconstruction Magnetic Resonance Imaging
Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, 2600 AA Delft, Netherlands

Oct 6, 2026J/K move · Enter open · S save

One Frame, Full Heartbeat: ECG-Free 4D Cardiac Cine MRI Synthesis via Radial-Decomposed Flow Matching

Shiyi Wang, Ruochen Sun, Xiang Li, Peirong Liu+1

Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA · Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA · Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA

Medical Image Synthesis Cardiac MRI 4D Reconstruction Physics-Informed Generative Modeling Flow-Based Generative Modeling

Open paper AskPDF arXiv

Cine cardiovascular magnetic resonance (CMR) captures the cardiac cycle as a four-dimensional (4D) sequence, but standard acquisition requires electrocardiogram (ECG) gating and repeated breath holds. Visual realism alone does not establish accurate patient-specific ejection fraction (EF) or ventricular volumes. We present PhaseFlow3D, a generative framework that synthesizes a complete 4D cine sequence from a single end-diastolic (ED) three-dimensional (3D) volume without ECG. To capture asymmetric systolic and diastolic dynamics, it represents the cardiac cycle as a piecewise linear phase anchored at ED and end-systolic (ES) time points. At inference, a population-level canonical template supplies this phase without patient-specific temporal information. A phase-conditioned rectified flow model generates a cardiac motion trajectory in latent space. Radial Contraction Decomposition converts each latent state into a 3D displacement field, combining a physics-informed radial component for centripetal myocardial contraction with an image-conditioned residual for rotation and out-of-plane motion. Each frame is generated by directly warping the ED volume, bypassing variational autoencoder decoding. On the combined ACDC and M&Ms benchmark, PhaseFlow3D achieves the lowest EF mean absolute error, the only positive left-ventricular volume-curve

R^2

, and the best distributional quality among compared methods. Ablations confirm each component's contribution. Downstream evaluations demonstrate the utility of the synthesized sequences and displacement fields for segmentation, pathology classification, label propagation, and myocardial strain analysis.

Abstract

Medical Image Synthesis

Explore similar work

One Frame, Full Heartbeat: ECG-Free 4D Cardiac Cine MRI Synthesis via Radial-Decomposed Flow Matching

ICHOR: A Robust Representation Learning Approach for ASL CBF Maps with Self-Supervised Masked Autoencoders

One Evaluation, Any Operating Point: Hypernetwork-Amortized MeanFlow for 3D MRI Reconstruction