现代电子技术2024，Vol.47Issue(7)：1-7,7.DOI:10.16652/j.issn.1004-373x.2024.07.001
USformer-Net:基于U-Net和Swin Transformer的脑部MRI图像质量评价方法
USformer-Net:brain MRI image quality assessment fusing U-Net and Swin Transformer
摘要
Abstract
Since the existing methods for assessing the quality of brain MRI(magnetic resonance imaging)images have low accuracy and it is difficult to apply them in the actual clinical environment,a brain MRI image quality automatic assessment model based on extraction of region of interest(ROI),named USformer-Net,is proposed.Additionally,a brain MRI image dataset with subjective quality assessment labels is established.The USformer-Net model is constructed based on combination of U-Net model and Swin Transformer model,with specific adaptations tailored to the characteristics of brain MRI images.The light U-Net network is used to segment the main brain regions which is of clinical diagnostic significance and extract the ROI.The Swin Transformer backbone incorporates operations of window-based multi-headed self-attention(W-MSA)and shifted-window multi-head self-attention(SW-MSA),as well as their feature fusion methods.Additionally,it integrates feature pyramid network(FPN),region of interest(ROI)align and a fully-connected network(FC)for image quality assessment.The USformer-Net model can effectively disregard irrelevant noise,accurately extract the critical regions affecting diagnosis,and perform image quality assessment.Experimental results demonstrate that the proposed model achieves accuracy of 87.84%,precision of 91.84%,recall rate of 92.05%,and F1-score of 91.99%in the task of MRI image quality assessment.In comparison with the other assessment methods,all indicators of the proposed model have been improved at varying degrees.The final results show that the model can effectively ensure the accuracy of brain MRI image quality assessment,and the created dataset with subjective quality assessment labels also provides better data support for the research in this field.关键词
图像质量评价/脑部MRI图像/深度学习/图像分割/U-Net/TransformerKey words
image quality assessment/brain MRI image/deep learning/image segmentation/U-Net/Transformer分类
电子信息工程引用本文复制引用
李沛钊,王同罕,贾惠珍,吴通..USformer-Net:基于U-Net和Swin Transformer的脑部MRI图像质量评价方法[J].现代电子技术,2024,47(7):1-7,7.基金项目
国家自然科学基金资助项目(62266001) （62266001）
国家自然科学基金资助项目(62261001) （62261001）
