Multi-objective Automatic Segmentation Model in MRI Images of Female Pelvic Floor Muscles
JIANG Wei
WANG Yan
Abstract：Aiming at the complex structure of female pelvic floor muscles and the small targets,which made it difficult to segment the target muscles in the images quickly and accurately,we proposed a new method for the automatic segmentation of female pelvic floor muscles in magnetic resonance imaging(MRI)images based on an improved attention u-shaped network(U-Net)structure,referred to as multi-scale convolutional block attention module U-Net(MCAtt-UNet).The model utilized a multi-scale CBAM(MCBAM)to capture richer image features,with its multi-scale characteristics being more advantageous for extracting features of small target muscles.Additionally,by embedding global context blocks between the encoding and decoding processes,the model effectively utilized contextual information and better captured global features.Experimental validation was conducted on a dataset provided by a medical university in Chongqing,including MRI images of pelvic floor muscles from 49 female subjects.The segmentation performance was evaluated using three metrics:Dice similarity coefficient(Dice),pixel accuracy(PA),and intersection over union(IOU).The achieved results for Dice,PA,and IOU were 76.03%,75.76%,and 64.15%,respectively,demonstrating overall superior segmentation performance compared to other networks.The results confirmed that the proposed model offers an alternative solution for multi-target segmentation of female pelvic floor muscles,aiding in the rapid and accurate diagnosis and treatment of clinically relevant pelvic floor disorders.
Keywords：female pelvic floor musclesMRI imagesconvolutional neural networksattention U-Netmulti-scale CBAM
Publication Date：2024-06-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 191-197 )