Structural MRI Harmonization via Disentangled Latent Energy-Based Style Translation.

Structural MRI Harmonization via Disentangled Latent Energy-Based Style Translation.
复制标题

通过解开的基于潜在能量的风格转换实现结构 MRI 协调。

DOI:
10.1007/978-3-031-45673-2_1
复制
发表时间:
2023
期刊:
Machine learning in medical imaging. MLMI (Workshop)
影响因子:
--
通讯作者:
Liu,Mingxia
Liu,Mingxia
中科院分区:
--
文献类型:
--
作者:
Wu,Mengqi;Zhang,Lintao;Yap,Pew-Thian;Lin,Weili;Zhu,Hongtu;Liu,Mingxia

文献摘要

相似文献

多位点脑磁共振成像(MRI)已广泛应用于临床和研究领域,但通常对由位点效应(如场强和扫描方案)引起的非生物学变化很敏感。几种回顾性数据协调方法在特征或全图像水平上消除这些非生物变异方面显示出有希望的结果。大多数现有的图像级协调方法都是通过生成对抗网络实现的,这种方法通常计算成本高,并且在独立数据上泛化能力差。为此,本文提出了一种基于解纠缠潜能(disentangled latent energy-based style translation,简称d唯恐)的图像级结构MRI协调框架。具体来说,通过潜在的自编码器和基于能量的模型,dleast解纠缠不变的图像生成和特定地点的风格翻译。自动编码器学习将图像编码到低维隐空间中,并从隐码中生成忠实图像。基于能量的模型被置于编码和生成步骤之间,促进了从源域到目标域的隐式风格转换。这允许高度泛化的图像生成和通过潜在空间的有效风格转换。我们在4092张t1加权mri上训练了我们的模型,分为3个任务:直方图比较、采集位点分类和脑组织分割。定性和定量结果证明了我们方法的优越性,它通常优于几种最先进的方法。
Multi-site brain magnetic resonance imaging (MRI) has been widely used in clinical and research domains, but usually is sensitive to non-biological variations caused by site effects (e.g., field strengths and scanning protocols). Several retrospective data harmonization methods have shown promising results in removing these non-biological variations at feature or whole-image level. Most existing image-level harmonization methods are implemented through generative adversarial networks, which are generally computationally expensive and generalize poorly on independent data. To this end, this paper proposes a disentangled latent energy-based style translation (DLEST) framework for image-level structural MRI harmonization. Specifically, DLEST disentanglessite-invariant image generationandsite-specific style translationvia a latent autoencoder and an energy-based model. The autoencoder learns to encode images into low-dimensional latent space, and generates faithful images from latent codes. The energy-based model is placed in between the encoding and generation steps, facilitating style translation from a source domain to a target domain implicitly. This allowshighly generalizable image generation and efficient style translationthrough the latent space. We train our model on 4,092 T1-weighted MRIs in 3 tasks: histogram comparison, acquisition site classification, and brain tissue segmentation. Qualitative and quantitative results demonstrate the superiority of our approach, which generally outperforms several state-of-the-art methods.