华中科技大学学报（自然科学版）2025，Vol.53Issue(11)：28-35,8.DOI:10.13245/j.hust.251137
面向机器人前列腺穿刺导航的图像分割网络
Image segmentation network for robot-assisted prostate biopsy navigation
摘要
Abstract
To address time-consuming and labor-intensive manual prostate segmentation in medical images,the limited effectiveness of existing automatic segmentation methods in fusing local and global information,and the reduced segmentation accuracy due to insufficient sensitivity to low-response regions,a segmentation network called CM-ProSegNet,combining convolutional neural networks(CNN)and multi-scale multi-layer perceptrons(MLP),was proposed,which was applied to segmentation of magnetic resonance imaging(MRI)and transrectal ultrasound(TRUS)images in a robotic prostate biopsy navigation system.In this network,multi-scale local semantic features were extracted by CNN and integrated into a set of feature maps,which were globally processed by MLP layers.Meanwhile,a compensation attention module was introduced to enhance the network's sensitivity to low-response regions,and to promote information fusion between the encoder and the decoder.Experimental results show that CM-ProSegNet achieves an average dice similarity coefficient(MDSC)of 91.98%and a Hausdorff distance(MHD)of 3.83 mm in magnetic resonance imaging(MRI)prostate segmentation;in TRUS segmentation,it achieves the MDSC of 94.42%and the MHD of 2.83 mm,outperforming existing mainstream segmentation methods.关键词
医学图像分割/穿刺机器人/前列腺/卷积神经网络/多层感知器Key words
medical image segmentation/biopsy robot/prostate/convolutional neural network/multi-layer perceptron分类
信息技术与安全科学引用本文复制引用
WANG Weirong,XU Kunning,LIU Yanjie,GU Jianjun..面向机器人前列腺穿刺导航的图像分割网络[J].华中科技大学学报（自然科学版）,2025,53(11):28-35,8.基金项目
国家自然科学基金资助项目(92048202) （92048202）
浙江省尖兵研发攻关计划资助项目(2023C03010). （2023C03010）
