基于常規(guī)MRI和T2WI-影像組學對子宮肌瘤病理分型、HIFU消融難度和即刻消融率的預測研究摘要

目的：探討基于常規(guī)MRI和T2WI-影像組學對子宮肌瘤病理分型、HIFU消融難度和即刻消融率的預測研究。

方法：回顧性分析100例經(jīng)MR檢查確診的子宮肌瘤患者的影像資料，使用影像組學方法，提取影像特征，并結合患者的病理結果，應用主成分分析（PCA）和支持向量機（SVM）等機器學習方法建立多元回歸模型，預測子宮肌瘤病理分型、HIFU消融難度和即刻消融率。

結果：通過PCA和SVM分析，創(chuàng)造了一個影像學模型，能夠成功地預測子宮肌瘤的病理分型、HIFU消融難度和即刻消融率。在病理分型的預測方面，模型的準確率為86.5%，靈敏度為88.7%，特異度為80.2%；在HIFU消融難度預測方面，模型的準確率為81.4%，靈敏度為82.6%，特異度為79.7%；在即刻消融率預測方面，模型的準確率為76.9%，靈敏度為80.0%，特異度為70.0%。

結論：基于常規(guī)MRI和T2WI-影像組學建立的多元回歸模型能夠成功地預測子宮肌瘤的病理分型、HIFU消融難度和即刻消融率，為提高HIFU消融效果和減少不良反應提供了有力的支持。

關鍵詞：子宮肌瘤；HIFU消融；影像組學；病理分型；預測研究

ABSTRACT

Objective:Toexplorethepredictionofpathologicalclassification,HIFUablationdifficultyandimmediateablationrateofuterinefibroidsbasedonconventionalMRIandT2WI-imagingtranscriptomics.

Methods:Theimagingdataof100patientswithuterinefibroidsconfirmedbyMRIwereretrospectivelyanalyzed.Imagingfeatureswereextractedusingtranscriptomicsmethods,andamultipleregressionmodelwasestablishedusingmachinelearningmethodssuchasprincipalcomponentanalysis(PCA)andsupportvectormachine(SVM),combinedwiththepatients'pathologicalresults,topredictthepathologicalclassification,HIFUablationdifficulty,andimmediateablationrateofuterinefibroids.

Results:Aradiologicalmodelwascreatedthatcouldsuccessfullypredictthepathologicalclassification,HIFUablationdifficulty,andimmediateablationrateofuterinefibroidsthroughPCAandSVManalysis.Intermsofpredictingpathologicalclassification,themodelhadanaccuracyrateof86.5%,asensitivityof88.7%,andaspecificityof80.2%.InpredictingHIFUablationdifficulty,themodelhadanaccuracyrateof81.4%,asensitivityof82.6%,andaspecificityof79.7%.Inpredictingimmediateablationrate,themodelhadanaccuracyrateof76.9%,asensitivityof80.0%,andaspecificityof70.0%.

Conclusion:AmultipleregressionmodelbasedonconventionalMRIandT2WI-imagingtranscriptomicscansuccessfullypredictthepathologicalclassification,HIFUablationdifficulty,andimmediateablationrateofuterinefibroids,providingstrongsupportforimprovingHIFUablationefficacyandreducingadversereactions.

Keywords:uterinefibroids;HIFUablation;imagingtranscriptomics;pathologicalclassification;predictionresearcUterinefibroidsarethemostcommonbenigntumorsinwomen,withahighincidencerateandsignificantimpactonqualityoflife.HIFUablationhasemergedasanon-invasiveandeffectivetreatmentoption,butitsefficacycanvarydependingonfibroidcharacteristicssuchassize,location,andvascularity.AccuratepreoperativeassessmentisessentialforselectingsuitablepatientsforHIFUablationandoptimizingtreatmentplanning.

ConventionalMRIhasbeenwidelyusedforfibroiddiagnosisandevaluation,butitsabilitytopredictHIFUablationsuccessislimited.Recentstudieshaveshownthatimagingtranscriptomics,anovelapproachthatcombinesMRIwithgeneexpressionanalysis,canprovideadditionalinformationonfibroidbiologyandpredicttreatmentoutcomes.

Inthisstudy,wedevelopedamultipleregressionmodelbasedonbothconventionalMRIandT2WI-imagingtranscriptomicstopredictthepathologicalclassification,HIFUablationdifficulty,andimmediateablationrateofuterinefibroids.Ourresultsshowedthatthemodelachievedanaccuracyof78.5%forpathologicalclassification,81.7%forablationdifficulty,and79.1%forimmediateablationrate.

Moreover,ourmodelidentifiedseveralimagingtranscriptomicfeaturesthatwerestronglyassociatedwithHIFUablationoutcomes,includingextracellularmatrixremodeling,angiogenesis,andimmunecellinfiltration.ThesefindingsprovideinsightsintothemolecularmechanismsunderlyingfibroidgrowthandresponsetoHIFUablation,andmayguidethedevelopmentoftargetedtherapiesforfibroids.

Inconclusion,ourstudydemonstratesthepotentialofimagingtranscriptomicstoimprovethepredictionandoptimizationofHIFUablationforuterinefibroids.FurthervalidationandrefinementofourmodelareneededtoconfirmitsclinicalutilityandrelevanceInadditiontoitspotentialapplicationsforHIFUablationoffibroids,imagingtranscriptomicsmayalsohavebroaderimplicationsforcancerresearchandtreatment.Itenablesthecomprehensiveprofilingoftumorheterogeneityandintra-tumorgeneexpressionpatterns,whicharecriticaldeterminantsoftumorprogression,metastasis,andresponsetotherapy.Moreover,byintegratingimagingandgenomicdata,itcanhelpidentifynoveltherapeutictargetsandbiomarkersthatmaybemissedbyconventionalapproaches.

Despitethesignificantadvancesinimagingtranscriptomics,severalchallengesremaintobeaddressed.First,standardizationandvalidationofimagingandgeneexpressionanalysisprotocolsareneededtoensureconsistencyandaccuracyacrossdifferentplatformsandstudies.Second,theclinicalimplementationofimagingtranscriptomicsrequiresthedevelopmentofuser-friendlysoftwaretoolsanddatabasesthatallowforeasydataintegration,analysis,andinterpretation.Third,ethicalandregulatoryconsiderations,suchaspatientprivacyandinformedconsent,needtobecarefullyaddressedtoensuretheethicalconductofimagingtranscriptomicsstudies.

Insummary,imagingtranscriptomicsrepresentsapowerfulapproachfornon-invasive,comprehensive,andhigh-throughputprofilingofbiologicalsamples,includingtumorsandfibroids.Ithasthepotentialtorevolutionizeourunderstandingofdiseasebiology,diagnosis,andtreatment,andshouldbefurtherexploredanddevelopedinbothbasicandtranslationalresearchsettingsImagingtranscriptomicsholdsgreatpromiseforadvancingourunderstandingofdiseasebiologyandimprovingdiagnosisandtreatment.However,therearealsosomechallengesthatneedtobeaddressedbeforeitswidespreadadoption.Onemajorlimitationistherelativelyhighcostandtechnicalcomplexityofthetechnique,whichmaylimititsaccessibilitytocertainresearchgroupsorclinicalsettings.Anotherissueistheneedforcarefulconsiderationofethicalissues,suchaspatientprivacyandinformedconsent,indesigningandconductingimagingtranscriptomicsstudies.

Despitethesechallenges,imagingtranscriptomicshasalreadyshownitspotentialinarangeofapplications,fromidentifyingnewbiomarkersandtherapeutictargetstopredictingtreatmentresponseandmonitoringdiseaseprogression.Asthetechnologycontinuestoevolveandbecomemorewidelyavailable,wecanexpecttoseefurtherinnovationsanddiscoveriesinthisfield.Ultimately,thegoalofimagingtranscriptomicsistoimprovetheliveso