Lightweight multimodal brain tumor MRI segmentation network for efficient clinical deployment: MLG-UNet achieves an average Dice of 86.67% with 0.81 M parameters

Zhang Xingpeng; Li Wei; Liu Xiaopeng; Sun Xinhuan; Wang Qiuli

doi:10.16016/j.2097-0927.202606058

您当前的位置：

首页 >

文章列表页 >

Lightweight multimodal brain tumor MRI segmentation network for efficient clinical deployment: MLG-UNet achieves an average Dice of 86.67% with 0.81 M parameters

Intelligent Medicine and Prediction Model | 更新时间：2026-10-30

- Lightweight multimodal brain tumor MRI segmentation network for efficient clinical deployment: MLG-UNet achieves an average Dice of 86.67% with 0.81 M parameters
- Journal of Army Medical University Vol. 48, Issue 20, Pages: 2920-2932(2026)
- 作者机构：
  
  1.西南石油大学计算机与软件学院，四川成都
  2.川北医学院核医药与辐射安全防控四川省重点实验室，四川南充
  3.渝粤病理科学研究中心金凤实验室，重庆
  4.陆军军医大学（第三军医大学）第一附属医院放射科，重庆
- 作者简介：
  
  Wang Qiuli，E-mail：wangqiuli@tmmu.edu.cn
- 基金信息：
  
  the Special Project of National Natural Science Foundation of China(62441610);the Joint General Project of Chongqing Municipal Science and Technology Commission(2025MSXM016);the Special Project for the Improvement of Science and Technology Innovation Capability of Army Medical University(2023XQN31)
- DOI：10.16016/j.2097-0927.202606058
  CLC： R319;R445.2;R739.41
- Received：12 June 2026，
  
  Revised：2026-09-04，
  
  Accepted：06 September 2026，
  
  Published：30 October 2026
- 稿件说明：
移动端阅览
张兴鹏，李伟，刘晓鹏，等.面向高效临床部署的轻量化多模态脑肿瘤MRI分割模型：MLG-UNet以0.81 M参数实现86.67%平均Dice［J］.陆军军医大学学报，2026，48（20）：2920-2932.DOI：10.16016/j.2097-0927.202606058.

Zhang X P，Li W，Liu X P，et al.Lightweight multimodal brain tumor MRI segmentation network for efficient clinical deployment： MLG-UNet achieves an average Dice of 86.67% with 0.81 M parameters［J］.J Army Med Univ，2026，48（20）：2920-2932.DOI：10.16016/j.2097-0927.202606058.
张兴鹏，李伟，刘晓鹏，等.面向高效临床部署的轻量化多模态脑肿瘤MRI分割模型：MLG-UNet以0.81 M参数实现86.67%平均Dice［J］.陆军军医大学学报，2026，48（20）：2920-2932.DOI：10.16016/j.2097-0927.202606058. DOI：

Zhang X P，Li W，Liu X P，et al.Lightweight multimodal brain tumor MRI segmentation network for efficient clinical deployment： MLG-UNet achieves an average Dice of 86.67% with 0.81 M parameters［J］.J Army Med Univ，2026，48（20）：2920-2932.DOI：10.16016/j.2097-0927.202606058. DOI：

摘要

目的

针对脑肿瘤MRI分割模型在影像科及临床科室端部署时面临的模型冗余度高、推理延迟高以及计算资源需求高等问题，探索在尽可能降低模型参数量和计算开销的同时保持较高分割精度的方法，以满足普通临床终端和资源受限设备的高效辅助分析需求。

方法

基于国际多模态脑肿瘤分割挑战赛（Brain Tumor Segmentation，BraTS）提供的BraTS2020与BraTS2019数据集开展研究，仅使用官方训练病例，按患者级别划分为训练集（BraTS2020：277例，BraTS2019：251例）、验证集（BraTS2020：37例，BraTS2019：34例）和测试集（BraTS2020：55例，BraTS2019：50例）后再提取二维轴向切片。提出轻量化多模态脑肿瘤MRI分割模型MLG-UNet，采用共享编码器-解码器结构，并设计分支上下文感知多核模块（branch-context multi-kernel block，BCMB）、低秩双线性融合模块（low-rank bilinear fusion，LRBF）和轻量级全局瓶颈语义调制模块（lightweight global bottleneck modulator，LGBM），分别增强多尺度特征提取、跨层特征交互和高层语义建模能力。将该模型与U-Net、TransUNet、UNeXt、EMCAD、Swin-Unet等代表性方法进行对比，并通过消融实验、可视化分析及患者级5轮重复留出验证全面评价模型性能与稳定性。

结果

在BraTS2020测试集上，MLG-UNet以0.81 M参数和0.53 G浮点运算次数（floating-point operations，FLOPs）实现86.67%的平均Dice和3.61 mm的平均第95百分位豪斯多夫距离（95th percentile Hausdorff distance，HD95），平均指标均优于全部对比方法；整体肿瘤区域、肿瘤核心区域、增强肿瘤区域子区域的Dice分别达到91.50%、88.82%和79.70%，对应HD95分别为3.37、3.58和3.89 mm。与平均Dice次优的TransUNet相比，MLG-UNet平均Dice提升0.49个百分点，而参数量由105.17 M降至0.81 M；与轻量方法UNeXt相比，在参数量和计算复杂度更低的情况下平均Dice提升0.69个百分点，平均HD95降低0.38 mm。在BraTS2019测试集上，平均Dice为86.51%、平均HD95为3.68 mm，同样优于全部对比方法，表明该模型具有较好的跨数据集泛化能力。消融实验表明BCMB、LRBF和LGBM均能有效提升分割性能；可视化比较显示本研究方法的分割结果与专家标注具有更高的重合度；患者级5轮重复留出验证中，平均Dice为（86.71±0.52）%，平均HD95为（3.65±0.14）mm，体现了较高的稳定性。

结论

MLG-UNet仅以0.81 M参数即实现86.67%的平均Dice，在控制模型规模和计算开销的同时也能够保持较高的脑肿瘤MRI分割精度，并在不同患者划分下表现出较稳定的分割性能。

Abstract

Objective

To address the problems of high model redundancy

high inference latency

and high computational resource demands faced by brain tumor MRI segmentation models when deployed in radiology departments and clinical settings

this study explores methods to maintain high segmentation accuracy while minimizing model parameters and computational overhead as much as possible

so as to meet the needs of efficient assisted analysis on general clinical terminals and resource-constrained devices.

Methods

Based on the BraTS2020 and BraTS2019 datasets from the international multimodal Brain Tumor Segmentation (BraTS) challenge

only the official training cases were used and divided at the patient level into training sets (BraTS2020: 277 cases; BraTS2019: 251 cases)

validation sets (BraTS2020: 37 cases; BraTS2019: 34 cases)

and test sets (BraTS2020: 55 cases; BraTS2019: 50 cases)

after which 2D axial slices were extracted. A lightweight multimodal brain tumor MRI segmentation model

termed MLG-UNet

was proposed

which adopted a shared encoder-decoder architecture and incorporates 3 designed modules: branch-context multi-kernel block (BCMB)

low-rank bilinear fusion module (LRBF)

and lightweight global bottleneck modulator (LGBM)

which enhance multi-scale feature extraction

cross-layer feature interaction

and high-level semantic modeling

respectively. The model was compared with representative methods

including U-Net

TransUNet

UNeXt

EMCAD

Swin-Unet

and model performance and stability were comprehensively evaluated through ablation experiments

visualization analysis

and patient-level five-round repeated holdout validation.

Results

On the BraTS2020 test set

MLG-UNet achieved a mean Dice of 86.67% and a mean 95th-percentile Hausdorff distance (HD95) of 3.61 mm with only 0.81 M parameters and 0.53 G floating-point operations (FLOPs)

and its mean metrics were superior to all compared methods. The Dice values for the whole tumor (WT)

tumor core (TC)

and enhancing tumor (ET) subregions reached 91.50%

88.82%

and 79.70%

with corresponding HD95 values of 3.37

3.58

and 3.89 mm

respectively. Compared with TransUNet

which achieved the second-best mean Dice

MLG-UNet improved the mean Dice by 0.49 percentage points

while reducing the parameters from 105.17 M to 0.81 M. Compared with the lightweight method UNeXt

MLG-UNet improved the mean Dice by 0.69 percentage points and reduced the mean HD95 by 0.38 mm

with lower parameter count and computational complexity. On the BraTS2019 test set

the mean Dice was 86.51% and the mean HD95 was 3.68 mm

also outperforming all compared methods

demonstrating favorable cross-dataset generalization. Ablation experiments showed that BCMB

LRBF

and LGBM all effectively improved segmentation performance. Visual comparison demonstrated higher agreement between the segmentation results of the proposed method and expert annotations. In patient-level five-round repeated holdout validation

the mean Dice was (86.71±0.52)% and the mean HD95 was (3.65±0.14) mm

indicating high stability.

Conclusion

MLG-UNet achieved a mean Dice of 86.67%

with only 0.81 M parameters

maintaining high brain tumor MRI segmentation accuracy while controlling model size and computational overhead

and exhibits stable segmentation performance across different patient partitions.

关键词

Keywords

references

吴进旭，吴云 . 残差混合注意力与自适应特征融合的脑肿瘤分割［J］. 计算机应用研究， 2025 ， 42 （ 8 ）： 2525 - 2531 . DOI： 10.19734/j.issn.1001-3695.2024.10.0432 http://dx.doi.org/10.19734/j.issn.1001-3695.2024.10.0432 .

Wu J X ， Wu Y . Brain tumor segmentation based on residual mixed attention and adaptive feature fusion ［J］. Appl Res Comput ， 2025 ， 42 （ 8 ）： 2525 - 2531 . DOI： 10.19734/j.issn.1001-3695.2024.10.0432 http://dx.doi.org/10.19734/j.issn.1001-3695.2024.10.0432 .

张兴鹏，刘晓鹏，李伟，等 . 融合显式边界先验与解耦注意力的多模态脑肿瘤分割网络［J/OL］. 中国图象图形学报， 2026 ， 1 - 18 . DOI： 10.11834/jig.260306 http://dx.doi.org/10.11834/jig.260306 .

Zhang X P ， Liu X P ， Li W ， et al . Multimodal brain tumor segmentation network fusing explicit boundary prior and decoupled attention ［J/OL］. J Image Graph ， 2026 ， 1 - 18 . DOI： 10.11834/jig.260306 http://dx.doi.org/10.11834/jig.260306 .

Menze B H ， Jakab A ， Bauer S ， et al . The multimodal brain tumor image segmentation benchmark （BRATS）［J］. IEEE Trans Med Imaging ， 2015 ， 34 （ 10 ）： 1993 - 2024 . DOI： 10.1109/tmi.2014.2377694 http://dx.doi.org/10.1109/tmi.2014.2377694 .

Zhang K ， Ganjizadeh A ， Vahdati S ， et al . Recent advances in compact portable platforms and gradient hardware for brain MRI ［J］. Radiology ， 2025 ， 315 （ 3 ）： e241904 . DOI： 10.1148/radiol.241904 http://dx.doi.org/10.1148/radiol.241904 .

Kimberly W T ， Sorby-Adams A J ， Webb A G ， et al . Brain imaging with portable low-field MRI ［J］. Nat Rev Bioeng ， 2023 ， 1 （ 9 ）： 617 - 630 . DOI： 10.1038/s44222-023-00086-w http://dx.doi.org/10.1038/s44222-023-00086-w .

陈锡程，卫泽良，叶伟，等 . 识别脑卒中微小病灶的多维度图像分割模型和系统开发［J］. 陆军军医大学学报， 2023 ， 45 （ 6 ）： 570 - 578 . DOI： 10.16016/j.2097-0927.202212169 http://dx.doi.org/10.16016/j.2097-0927.202212169 .

Chen X C ， Wei Z L ， Ye W ， et al . Multi-dimensional segmentation model and system development for recognizing small lesions after acute ischemic stroke ［J］. J Army Med Univ ， 2023 ， 45 （ 6 ）： 570 - 578 . DOI： 10.16016/j.2097-0927.202212169 http://dx.doi.org/10.16016/j.2097-0927.202212169 .

Roy S ， Koehler G ， Ulrich C ， et al . MedNeXt： transformer-driven scaling of ConvNets for medical image segmentation ［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2023 ， 2023 ： 405 - 415 . DOI： 10.1007/978-3-031-43901-8_39 http://dx.doi.org/10.1007/978-3-031-43901-8_39 .

Wang H N ， Cao P ， Yang J Z ， et al . Narrowing the semantic gaps in U-Net with learnable skip connections： The case of medical image segmentation ［J］. Neural Netw ， 2024 ， 178 ： 106546 . DOI： 10.1016/j.neunet.2024.106546 http://dx.doi.org/10.1016/j.neunet.2024.106546 .

Lu Z K ， She C Y ， Wang W ， et al . LM-Net： a light-weight and multi-scale network for medical image segmentation ［J］. Comput Biol Med ， 2024 ， 168 ： 107717 . DOI： 10.1016/j.compbiomed.2023.107717 http://dx.doi.org/10.1016/j.compbiomed.2023.107717 .

Rahman M M ， Marculescu R . MK-UNet： multi-kernel lightweight CNN for medical image segmentation ［C］// 2025 IEEE/CVF International Conference on Computer Vision Workshops （ICCVW）， 2025 ： 1053 - 1062 . DOI： 10.1109/iccvw69036.2025.00114 http://dx.doi.org/10.1109/iccvw69036.2025.00114 .

Xing Z H ， Ye T ， Yang Y J ， et al . SegMamba： long-range sequential modeling mamba for 3D medical image segmentation ［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2024 . Cham ： Springer Nature Switzerland ， 2024 ： 578 - 588 . DOI： 10.1007/978-3-031-72111-3_54 http://dx.doi.org/10.1007/978-3-031-72111-3_54 .

Ruan J C ， Li J C ， Xiang S C . VM-UNet： vision mamba UNet for medical image segmentation ［J］. ACM Trans Multimedia Comput Commun Appl ， 2025 ： 3767748 . DOI： 10.1145/3767748 http://dx.doi.org/10.1145/3767748 .

Ronneberger O ， Fischer P ， Brox T . U-Net： convolutional networks for biomedical image segmentation ［C］// Medical Image Computing and Computer-Assisted Intervention—MICCAI 2015 . Cham ： Springer International Publishing ， 2015 ： 234 - 241 . DOI： 10.1007/978-3-319-24574-4_28 http://dx.doi.org/10.1007/978-3-319-24574-4_28 .

Xie E Z ， Wang W H ， Yu Z D ， et al . SegFormer： simple and efficient design for semantic segmentation with transformers ［C］// Advances in Neural Information Processing Systems（NeurIPS）， 2021 ： 12077 - 12090 . DOI： 10.48550/arXiv.2105.15203 http://dx.doi.org/10.48550/arXiv.2105.15203 .

Huang X H ， Deng Z F ， Li D D ， et al . MISSFormer： an effective transformer for 2D medical image segmentation ［J］. IEEE Trans Med Imaging ， 2023 ， 42 （ 5 ）： 1484 - 1494 . DOI： 10.1109/TMI.2022.3230943 http://dx.doi.org/10.1109/TMI.2022.3230943 .

Cao H ， Wang Y Y ， Chen J ， et al . Swin-unet： unet-like pure transformer for Medical image segmentation ［C］// Computer Vision—ECCV 2022 Workshops . Cham ： Springer Nature Switzerland ， 2023 ： 205 - 218 . DOI： 10.1007/978-3-031-25066-8_9 http://dx.doi.org/10.1007/978-3-031-25066-8_9 .

Chen J N ， Mei J R ， Li X H ， et al . TransUNet： Rethinking the U-Net architecture design for medical image segmentation through the lens of transformers ［J］. Med Image Anal ， 2024 ， 97 ： 103280 . DOI： 10.1016/j.media.2024.103280 http://dx.doi.org/10.1016/j.media.2024.103280 .

Valanarasu J M J ， Patel V M . UNeXt： MLP-based rapid medical image segmentation network ［C］// Medical Image Computing and Computer Assisted Intervention—MICCAI 2022 . Cham ： Springer Nature Switzerland ， 2022 ： 23 - 33 . DOI： 10.1007/978-3-031-16443-9_3 http://dx.doi.org/10.1007/978-3-031-16443-9_3 .

Rahman M M ， Munir M ， Marculescu R . EMCAD： efficient multi-scale convolutional attention decoding for medical image segmentation ［C］// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）， 2024 ： 11769 - 11779 . DOI： 10.1109/cvpr52733.2024.01118 http://dx.doi.org/10.1109/cvpr52733.2024.01118 .

Gui C Z ， An X W ， Liu S ， et al . S 2 Net： Self-adaptive weighted fusion and self-adaptive aligned network for multi-modal MRI segmentation ［J］. Med Phys ， 2025 ， 52 （ 6 ）： 4175 - 4187 . DOI： 10.1002/mp.17742 http://dx.doi.org/10.1002/mp.17742 .

Li H Z ， Ren Z H ， Zhu G Q ， et al . Enhancing medical image segmentation with MA-UNet： a multi-scale attention framework ［J］. Visual Comput ， 2025 ， 41 （ 8 ）： 6103 - 6120 . DOI： 10.1007/s00371-024-03774-9 http://dx.doi.org/10.1007/s00371-024-03774-9 .

Zhao L X ， Wang T ， Chen Y B ， et al . MSAByNet： a multiscale subtraction attention network framework based on Bayesian loss for medical image segmentation ［J］. Biomed Signal Process Control ， 2025 ， 103 ： 107393 . DOI： 10.1016/j.bspc.2024.107393 http://dx.doi.org/10.1016/j.bspc.2024.107393 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Automatic segmentation of female urine control anatomical elements and related structures in MRI images based on deep learning

Multi-dimensional segmentation model and system development for recognizing small lesions after acute ischemic stroke

Glioblastoma neovascularization and MRI evaluation

Advanced MR imaging in diagnosis of brain glioma: based on its metabolic and physiological characteristics

A 3 mL low-dose gadolinium-based contrast agent combined with 3D-TWIST sequence yields diagnostic accuracy equivalent to standard-dose CE-MRV in patients weighing <70 kg: a non-randomized controlled study

Related Author

ZHANG Ziqin

WU Yi

ZHANG Xiaoqin

XU Zhou

LEI Ling

WANG Yanzhou

WANG Yan

Xicheng CHEN

Related Institution

School of Mathematical Sciences， Chongqing Normal University

Department of Digital Medicine， Faculty of Biomedical Engineering and Imaging Medicine， Army Medical University （Third Military Medical University）

Department of Gynecology， Anshun People’s Hospital

Department of Obstetrics and Gynecology， First Affiliated Hospital， Army Medical University （Third Military Medical University）

Department of Health Statistics, Faculty of Military Preventive Medicine, Army Medical University (Third Military Medical University)

⁰