多特征融合的产时超声胎方位识别模型

打开文本图片集

A multi-feature fusion-based model for fetal orientation classification from intrapartum ultrasound videos

ZHENGZiyu，Giying，engjie，HGie，uorong，ezhong，NGJuntH 1DepartmentofforationostuctiondagemntColegeofnineeingHaioiesityQuazou; 3Departmntofscdfspiaoficleityonapmte andGynecology，AffiliatedPeople'sHospitalofQuanzhouMedical College，Quanzhou36218，China

Abstract：ObjectiveToconstructanintellgentanalysis modelforclasifyingfetalorientationduringintrapartumultrasound videosbasedonmulti-featurefusion.MethodsTheproposedmodelconsistsoftheInput，BackboneNetworkand Classification Head modules.The Inputmodulecariesout dataaugmentationtoimprovethesamplequalityand generalizationabilityofthemodel.TheBackboneNetwork wasresponsibleforfeatureextractionbasedonYolov8combined withCBAM，ECA，SAantionmechismandatureiteractionoule.lasificationHeadoistsofa convolutionallayerandasoftmaxfunctiontooutputthefinalprobabilityvalueofeachclas.Theimagesofthekeystructures （theeyes，face，ead，thalamusandspine）wereannotatedithframesbypysiciansformodeltrainingtoimprovethe clasificationaccuracyoftheanterioroipital，posteriorocipitalandtransverseoccipitalorientations.Resultshe experimentalresultsshowedthat theproposedmodelhadexcellentperformance inthetireorientationclasificationtask with theclasificationaccuracyreaching 0.984，anareaunder thePRcurve （averageaccracy）of 0.993，andarea under theROC curve of 0.984，anda kappa consistencytest scoreof 0.974.The prediction resultsbythe dep learning model were highly consistentwith theactual clasificationresults.ConclusionThemulti-feature fusionmodel proposed inthisstudycan efficientlyand accuratelyclassify fetal orientation in intrapartumultrasound videos.

Keywords： intrapartumultrasound; fetal orientation;deeplearning;attentionmechanism

产时超声（IPUS）起源于20世纪50年代，是一种在分娩过程中使用的超声技术，随着技术的发展进步，产时超声广泛地应用于产时临床1，妇产科超声学会（ISUOG）发布的产时超声检查指南中强调产时超声在促进分娩临床管理中的重要性，并提供正确使用超声指导分娩的建议。（剩余11881字）