跳转至

文章背景与核心概要

评估脑血管储备（CVR）对于烟雾病患者确定是否需要进行颅外-颅内搭桥手术至关重要。该评估通常需要配对的动脉自旋标记（ASL）灌注MRI扫描——一次在基线状态下，另一次在静脉注射乙酰唑胺（ACZ）之后。

在ACZ禁忌或需避免使用的病例中，无法获取注射ACZ后的脑血流（CBF）图。本文介绍了 CAE3D，这是一种确定性3D条件自编码器，旨在直接从ACZ注射前的ASL输入中合成注射ACZ后的CBF图。与各种2D和3D基线模型相比，该模型表现出更优越的性能，为无法耐受完整双扫描方案的患者进行回顾性血流动力学评估提供了一种潜在的途径。

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

arXiv: 2608.14758
Authors: Julia Huang, Camila Gonzalez, Rydham Goyal, Aja Zou, Sasha Alexander, Michael Moseley, Moss Y. Zhao, Gary K. Steinberg
Published: August 14, 2026
Venue: Accepted at Machine Learning for Healthcare (MLHC 2026)

arXiv: 2608.14758
Authors: Julia Huang, Camila Gonzalez, Rydham Goyal, Aja Zou, Sasha Alexander, Michael Moseley, Moss Y. Zhao, Gary K. Steinberg
Published: August 14, 2026
Venue: Accepted at Machine Learning for Healthcare (MLHC 2026)

Summary

对于烟雾病患者而言，评估脑血管储备（CVR）是决定是否进行颅外-颅内搭桥手术的关键步骤。这一评估通常需要成对的动脉自旋标记（ASL）灌注MRI扫描——一个在基线时进行，另一个在注射乙酰唑胺（ACZ）后进行。

For patients with Moyamoya disease, assessing cerebrovascular reserve (CVR) is a critical step in determining the necessity of extracranial-to-intracranial bypass surgery. This assessment typically requires paired arterial spin labeling (ASL) perfusion MRI scans—one at baseline and one post-acetazolamide (ACZ) administration.

在ACZ禁忌或需避免使用的情况下，无法获得注射ACZ后的脑血流（CBF）图。本文引入了 CAE3D，这是一种确定性的3D条件自编码器，旨在直接从ACZ预处理的ASL输入中合成ACZ后CBF图。与各种2D和3D基线相比，该模型表现出优越的性能，为无法接受完整双扫描方案的患者提供了一条潜在的回顾性血流动力学评估途径。

In cases where ACZ is contraindicated or avoided, the post-ACZ cerebral blood flow (CBF) map is unavailable. This paper introduces CAE3D, a deterministic 3D conditional autoencoder designed to synthesize post-ACZ CBF maps directly from pre-ACZ ASL inputs. The model demonstrates superior performance compared to various 2D and 3D baselines, offering a potential pathway for retrospective hemodynamic assessment in patients who cannot undergo the full two-scan protocol.

Key Findings

  • 模型性能： CAE3D在独立测试集上实现了0.066的平均绝对误差（MAE）、0.80的结构相似性指数（SSIM）以及24.0 dB的峰值信噪比（PSNR）。
  • 统计显著性： 在MAE指标上，该模型优于八个从头训练（trained-from-scratch）基线中的七个；在SSIM和PSNR指标上，则超越了全部八个基线。
  • 临床实用性： 该模型表现出接近零的全脑平均偏差，表明其用于临床合成具有高度可靠性。
  • 局限性： 尽管该模型对回顾性分析有效，但研究人员指出，要将其扩展到禁用ACZ的患者群体，还需要进一步的外部验证和前瞻性验证。
  • Model Performance: CAE3D achieved a held-out Mean Absolute Error (MAE) of 0.066, an SSIM of 0.80, and a PSNR of 24.0 dB.
  • Statistical Significance: The model outperformed seven of eight trained-from-scratch baselines in MAE and surpassed all eight baselines in SSIM and PSNR.
  • Clinical Utility: The model exhibits near-zero full-brain mean bias, suggesting high reliability for clinical synthesis.
  • Limitations: While effective for retrospective analysis, the researchers note that extension to patients where ACZ is contraindicated requires further external and prospective validation.

Accessing the Paper

Metadata