金字塔信息解耦的甲状腺结节分割

打开文本图片集

关键词：甲状腺结节分割；信息解耦；跨层级金字塔；样条变化

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2026）04-037-1266-06

doi：10.19734/j.issn. 1001-3695.2025.06.0262

Pyramid information decoupling for thyroid nodule segmentation

Zhang Ziyuan’，Lyu Xiaoqi1，2t （1.SchoolofigialellgedtreooliaUesitcedootouNooo lege of Information Engineering， Inner Mongolia University of Technology，Huhhot O1oo51，China）

Abstract：Thyroid nodule segmentation playsacrucial role indiagnosing thyroiddiseases，but limiteddatasetsandultrasound noiseconstrainitsaccuracy.Totacklethese，thispaperproposedapyamid-information-decoupling thyoidnodulesegmentation network.Itemployed multi-branchfusiontodecouplefeaturesandachieve cyclicstate transfer，therebyadapting tonodulecale variationsandprogressvelyrefiningstructuraldetails.Themodelintegratedacros-levelfeaturepyaidwithmultiplerossatention mechanismstofacilitatebidirectioalfeature interactionandsupressnoiseinterference.Italsointroducedasplinebasedrefinementmoduletodynamicallyadjusttoinputcontent，capturefinedetails，andimprovegeneralization.OnheT3K dataset，themodelachevedaverageIoUandDSCscoresofO.7633andO.8599，demonstratingthatitoutperforms mainstream methods.It alsoshowsoutstanding performance in generalizationscenariosonthesmal-sampleDDTIandTNUIdatasets，confirming its significant potential for thyroid nodule segmentation tasks.

Key words：thyroid nodule segmentation；information decoupling；cros-level pyramid；spline variation

0引言

甲状腺结节作为甲状腺内的异常肿块，其近30年的检出率显著上升。（剩余15699字）