EchoGPK基于先验知识引导的超声心动图轻量级图卷积分析方法
文档名：EchoGPK基于先验知识引导的超声心动图轻量级图卷积分析方法
摘要：根据超声心动图准确分析左心室轮廓和射血分数对于心血管疾病诊断意义重大.但现有方法存在左心室分割和射血分数预测之间缺乏关联性、左心室分割关键点易于出现离群点和突变点、方法存储和计算开销大、解释性不佳等问题,为此提出一种基于先验知识引导的轻量级图卷积方法EchoGPK(EchoGuidedbyPrioriKnowledge),以心脏的结构和运动特性、相邻心肌的相似性等先验知识为引导,设计了计算高效的螺旋聚合函数和深度压缩的多头偏心聚合解码器,实现了图卷积结构的轻量化.方法基于临床医生的普遍经验提出了适度利用左心室轮廓的多任务射血分数预测网络,建立了左心室分割和射血分数预测之间的关联性,增强了推理的可解释性;基于图卷积神经网络的传递特性约束邻居点的行为,减少了边界离群点和突变点的产生.EchoGPK在大型公开数据集EchoNet-Dynamic上的实验结果表明,左心室分割的Dice分数达92.13％,射血分数预测的MAE达3.92％;方法表现出准确率高、参数量和算力需求低等特点,证明了先验知识在超声医学图像分析中的有效性.
Abstract：Accurateanalysisoftheleftventricularoutlineandejectionfractionthroughechocardiographyholdssig-nificantdiagnosticimplicationsincardiovasculardiseases.However,currentmethodologiesexhibitdeficienciessuchasalackofcorrelationbetweenleftventricularsegmentationandejectionfractionprediction,susceptibilitytooutliersandabruptvariationsinkeypointsofleftventricularsegmentation,substantialstorageandcomputationaloverhead,andpoorin-terpretability.Inaddressingtheseissues,thisstudyproposesalightweightgraphconvolutionalmethodtermedEchoGPK(EchoGuidedbyPrioriKnowledge).Guidedbypriorknowledgeencompassingcardiacstructure,motioncharacteristics,andthesimilarityamongadjacentmyocardialregions,theapproachincorporatesacomputationallyefficientspiralaggrega-tionfunctionandadeeplycompressedmulti-headeccentricaggregationdecoder,achievingthelightweightingofthegraphconvolutionalstructure.Leveragingthecommonexperiencesofclinicalpractitioners,themethodintroducesamultitaskejectionfractionpredictionnetworkthatmoderatelyutilizesleftventricularcontours,establishingacorrelationbetweenleftventricularsegmentationandejectionfractionpredictiontoenhanceinterpretability.Byemployingthegraphconvolutionalneuralnetworktransmissioncharacteristicstoconstrainthebehaviorofneighboringpoints,thegenerationofboundaryoutli-ersandabruptvariationsisreduced.Experimentalresultsonthelarge-scalepublicdatasetEchoNet-DynamicdemonstratethatEchoGPKachievesaDicescoreof92.13％forleftventricularsegmentationandameanabsoluteerror(MAE)of3.92％forejectionfractionprediction.Furthermore,themethodexhibitshigheraccuracy,superiorparametercountandcomputa-tionalefficiencycomparedtorelevantapproaches,affirmingtheeffectivenessofpriorknowledgeinultrasoundmedicalim-ageanalysis.
作者：王博荣 叶剑 Author：WANGBo-rong YEJian
作者单位：中国科学院计算技术研究所,北京100190;中国科学院大学,北京101408中国科学院计算技术研究所,北京100190;移动计算与新型终端北京市重点实验室,北京100190
刊名：电子学报 ISTICEIPKU
Journal：ActaElectronicaSinica
年，卷(期)：2024, 52(4)
分类号：TP391.41
关键词：关键超声心动图左心室分割射血分数预测图卷积神经网络
Keywords：echocardiographyleftventricularsegmentationejectionfractionspredictiongraphconvolutionalneu-ralnetwork
机标分类号：TP391.41TN911.6R541.6
在线出版日期：2024年6月26日
基金项目：EchoGPK:基于先验知识引导的超声心动图轻量级图卷积分析方法[
期刊论文]电子学报--2024, 52(4)王博荣叶剑根据超声心动图准确分析左心室轮廓和射血分数对于心血管疾病诊断意义重大.但现有方法存在左心室分割和射血分数预测之间缺乏关联性、左心室分割关键点易于出现离群点和突变点、方法存储和计算开销大、解释性不佳等问题,为...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
EchoGPK:基于先验知识引导的超声心动图轻量级图卷积分析方法EchoGPK:A Lightweight Graph Convolutional Analysis Method for Echocardiography Based on Prior Knowledge Guidance
EchoGPK:基于先验知识引导的超声心动图轻量级图卷积分析方法.pdf
页:
[1]