Motion correction for brain MRI using deep learning and a novel hybrid loss function Lei Zhang1 Xiaoke Wang1 Michael Rawson2 Radu Balan3 Edward H. Herskovits1 Elias

2025-05-02 0 0 1.42MB 30 页 10玖币

侵权投诉

Motion correction for brain MRI using deep learning and a novel

hybrid loss function

Lei Zhang1, Xiaoke Wang1, Michael Rawson2, Radu Balan3, Edward H. Herskovits1, Elias

Melhem1, Linda Chang1, Ze Wang1, and Thomas Ernst1

1Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of

Medicine, Baltimore, MD, USA

2Pacific Northwest National Laboratory

3Department of Mathematics and Center for Scientific Computation and Mathematical Modeling,

University of Maryland, College Park, MD, USA,

Corresponding Authors

Thomas Ernst, Ph.D.

Department of Diagnostic Radiology and Nuclear Medicine

University of Maryland School of Medicine

670 W. Baltimore Street, HSF-III, Room 1130,

Baltimore, MD 21202

Email: ternst@som.umaryland.edu

Ze Wang, Ph.D.

Department of Diagnostic Radiology and Nuclear Medicine

University of Maryland School of Medicine

670 W. Baltimore Street, HSF-III, Room 1130,

Baltimore, MD 21202

Email: ze.wang@som.umaryland.edu

Figure and Tables: 9 Figures and 1 Table (1 supplementary table)

Key words: MRI, motion correction, deep learning, brain

Abstract

Purpose

To develop and evaluate a deep learning-based method (MC-Net) to suppress motion artifacts in

brain magnetic resonance imaging (MRI).

Methods

MC-Net was derived from a UNet combined with a two-stage multi-loss function. T1-weighted

axial brain images contaminated with synthetic motions were used to train the network.

Evaluation used simulated T1 and T2-weighted axial, coronal, and sagittal images unseen during

training, as well as T1-weighted images with motion artifacts from real scans. Performance

indices included the peak signal to noise ratio (PSNR), structural similarity index measure

(SSIM), and visual reading scores. Two clinical readers scored the images.

Results

The MC-Net outperformed other methods implemented in terms of PSNR and SSIM on the T1

axial test set. The MC-Net significantly improved the quality of all T1-weighted images (for all

directions and for simulated as well as real motion artifacts), both on quantitative measures and

visual scores. However, the MC-Net performed poorly on images of untrained contrast (T2-

weighted).

Conclusion

The proposed two-stage multi-loss MC-Net can effectively suppress motion artifacts in brain

MRI without compromising image context. Given the efficiency of the MC-Net (single image

processing time ~40ms), it can potentially be used in real clinical settings. To facilitate further

research, the code and trained model are available at https://github.com/MRIMoCo

/DL_Motion_Correction.

Introduction

Magnetic resonance imaging (MRI) is a widely used medical imaging modality due to its

ability of visualizing both the anatomy and function of tissues and organs as well as pathologic

processes1. MRI provides high spatial resolution and diverse contrasts, making it superior to

many other imaging modalities for detecting and characterizing soft tissue (e.g., brain, abdominal

organs, and blood vessels) and pathologies.

Because of the sequential spatial encoding steps used to spatially encode the imaging object,

MRI is relatively slow and can take up to several minutes for a typical 3D volume scan. The

prolonged image acquisition process makes MRI sensitive to motion1,2. Unfortunately, motion in

human subjects is inevitable and can be caused by involuntary physiological movements, such as

respiration and cardiac motion, and unintended patient movements. Motion-induced image

artifacts can drastically deteriorate image quality and reduce diagnostic accuracy2. For example,

Andre et al reported that almost 60% of 192 clinical brain MRI scans were contaminated with

motion artifacts3. Among these, 28% were marginally diagnostic to non-diagnostic and should be

repeated. Because of the motion-induced image artifacts, the annual loss of revenues per MR

scanner can be over $100,000 for brain studies alone3.

A range of prospective correction strategies have been developed for motion artifacts 4–7 8,

but they commonly have limitations such as scanner platform accessibility, applicability to

specific MR imaging sequences, and limitations in correcting different types of motion artifacts

(e.g., in-plane versus through-plane movements). Therefore, retrospective motion-correction by

means of post-processing provides a good complement. One promising approach involves deep

learning (DL)1,2,9–16,17,18, using deep convolutional neural networks (DCNN) or other network

architectures with supervised learning. Given sufficient training pairs (inputs and reference

images), DCNNs can be trained to learn the transformation from the input (motion-corrupted

image) to the reference (motion-free image). Trained DCNNs have been used successfully to

solve many challenging and clinically important problems, e.g., arterial spin labeling perfusion

MRI denoising 13,19, image segmentation 20,21, and image registration 22,23.

DCNNs appear to be well-suited for retrospective correction of motion artifacts since there

are no obvious conventional algorithms to solve the problem and yet expert readers can “read

through” the artifacts to some degree. Recent studies demonstrate that DCNNs can be used to

attenuate motion artifacts in brain MRI scans using a data-driven approach without prior

knowledge. For instance, variation auto encoders (VAE) and generative adversarial networks

(GAN) were implemented for retrospective correction of rigid and non-rigid motion artifacts

from motion-affected MR scans1. GANs were also used for motion correction in multi-shot MRI

12. A conditional GAN improved the image quality of predicted motion-corrected images

compared to motion-corrupted images24. Finally, an encoder-decoder network was able to

suppress motion artifacts with motion simulation augmentation2.

The purpose of this study was to implement and comprehensively evaluate a new deep

neural network architecture and loss function for motion correction. The methodology and scope

of this study are different from previous studies. The novel and unique contributions of this paper

include: First, a new loss function was proposed, which contained an L1 component for

penalizing overall image artifacts and a total variation component to penalize the loss of image

details such as boundaries. Accordingly, a two-stage training strategy was implemented to first

minimize the overall motion-artifacts and then consider both the residual motion-induced

artifacts and the loss of image details such as boundaries. Second, the generalizability of the

trained model was assessed using images with different contrast than that of the training data.

Third, to ensure rigor and demonstrate clinical utility, substantial evaluations were made using

different levels of synthetic motions and in-vivo data through both objective performance indices

and subjective reading by experienced clinicians. Motion-free images were also used to assess

potential over-corrections by the trained DL networks. Finally, to allow other researchers to

reproduce our work or use the methods to process their own data, we have released the code and

sample data at https://github.com/MRIMoCo/DL_Motion_Correction.

Methods

2.1 MC-Net

The proposed deep learning-based method (MC-Net) takes a motion‐corrupted image as

input and outputs a motion‐corrected image. The method implements a modified UNet (Figure 1)

as its neural-network structure. The MC-Net was trained with a novel two-stage training strategy

using a hybrid loss function L that combines a L1-loss and Total Variation loss (TV) 25:

         (1)

    

 (2)

         

 (3)

where I and I0 are a corrupted image and a motion free image, and i and j are row/column

indices. During the first training stage, we used the L1-loss only [(alpha, beta) = (1, 0)] to

suppress overall motion-induced artifacts. The pre-trained stage 1 model is then fine-tuned in

stage 2 by turning on the TV-loss component [(alpha, beta) = (1, 1)]; this penalizes boundary

artifacts in addition to the overall artifacts.

2.2 Motion Corrupted Images

The pipeline that generates motion-corrupted k-space data is shown in Figure 2. The project

used images with simulated motion artifacts, based on deidentified brain MRI scans from 52

human subjects (50 male, 2 female, age 48.6 ± 9.1 years) previously enrolled in research

studies. All data were acquired on a 3T scanner (TIM Trio, Siemens Healthcare, Erlangen,

Germany). The ability of the MC-Net to correct real (non-simulated) images with motion

artifacts was assessed using motion-corrupted scans from five additional subjects (2 male, 3

female, age 19±4.9).

The source images were 3D sagittal magnetization-prepared rapid gradient-echo (MP-

RAGE) scans and 2D axial fluid-attenuated inversion-recovery (FLAIR) scans from 52 subjects.

MP-RAGE images were collected with the following parameters: TR=2.2s; TE=4.47ms; TI=1s;

resolution=1mm isotropic; matrix size=256×256×160, and FLAIR images were collected with

the following parameters: TR=9.1s; TE=84ms; echo train length=11; matrix size=256×204; in-

plane resolution = 1mm2; slice thickness=3mm; slice spacing=3mm; TI=2.5s. All source images

were assessed visually to ensure they did not contain motion artifacts.

Forty-two axial in-plane motion trajectories of 256 temporal samples each were synthesized

from in-vivo head movements measured with the prospective acquisition correction (PACE)

algorithm27 during BOLD functional MRI (fMRI) scans. The source motion trajectories had

translations <2mm and rotations <2°, and were subsequently multiplied by eight and reduced

from 6 degrees of freedom to in-plane motions (3 degrees of freedom). The offsets of the

trajectories were normalized such that the motions in all axes are zero at the center of the

trajectory. The severity of each motion trajectory applied is indicated by the motion standard

deviation across time for all three in-plane degrees of freedom (L2-norm of in-plane translations

in mm and in-plane rotation in degree).

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

MotioncorrectionforbrainMRIusingdeeplearningandanovelhybridlossfunctionLeiZhang1,XiaokeWang1,MichaelRawson2,RaduBalan3,EdwardH.Herskovits1,EliasMelhem1,LindaChang1,ZeWang1,andThomasErnst11DepartmentofDiagnosticRadiologyandNuclearMedicine,UniversityofMarylandSchoolofMedicine,Baltimore,MD,USA2PacificN...

收起<<

Motion correction for brain MRI using deep learning and a novel hybrid loss function Lei Zhang1 Xiaoke Wang1 Michael Rawson2 Radu Balan3 Edward H. Herskovits1 Elias.pdf

共30页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Motion correction for brain MRI using deep learning and a novel hybrid loss function Lei Zhang1 Xiaoke Wang1 Michael Rawson2 Radu Balan3 Edward H. Herskovits1 Elias

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: