基于改进U-Net的细胞核图像分割网络

打开文本图片集

Anucleus image segmentation network based on improved U-Net

SONGWenbo，ZHU Kaiyan，LIU Tong，SONGWeibo （School of InformationEngineering，DalianOceanUniversity，Dalian116O23，China）

Abstract：Deep learning model basedonconvolutional neural network（CNN） haveachieved significantbreakthrough in biomedicalimagesegmentationandhavebeen widelyappliedinpracticalscenarios.Theaccuracyof nucleus imagesegmentation playsacrucialroleinpathologicaldiagnosis.However，theexistingnucleussegmentationalgorithsstillsuferfromisuessuch asfuzzyandadherentboundaries，soanimagesegmentationalgorihmbasedonimprovedU-Netisproposed.Inthemodel，a tripleattentionmoduleisutizedtoenhancefeature focus，andfeaturefusion module，AGgatemodule，andlightweightInception moduleareincorporatedtoimprovesegmentationaccuracy.Theproposedalgorithmwasvalidatedonthe2O18DataScienceBowl （DSB2018）dataset.The evaluationmetrics including IoU（intersectionover union）and DSC reach 81.85%and 90.00% respectively.Experimentalresultsdemonstratethatincomparisonwiththeothersegmentationmodels，theproposedalgorithm exhibitssignificantadvantagesintermsof theconformitybetweensegmentedresultsandgroundtruthlabels，achievingsuperior performance.

Keywords：CNN;deep learning;nucleus segmentation;U-Netnetwork;atentionmechanism;image segmentation

0 引言

细胞核图像分割作为医学图像分析的关键一环，扮演着鉴定细胞、了解生物过程以及提高药物检测效率的重要角色。（剩余10565字）