基于临床及影像学特征的机器学习模型诊断乳腺癌腋窝淋巴转移负荷的临床价值

打开文本图片集

ABSTRACTObjectiveTo constructa machine learning model based on clinical and imaging features，and to explore itsclinicalvalueindiagnosingaxillarylymphnodemetastasis（ALNM）burdeninbreastcancerpatientswithimaging-negative axillryymphodes.MethodsAtotalof473breastcancerpatientswhounderwentaxillrylymphnodebiopsyordssectionin ourhospital wereenroledanddividedintoatrainingset（8Ocases）andatestingset（93cases）witharatioof8：2.Dierencesin clinicaldata，ultrasoundfindings，andMRIresultsbetweenthetwosets werecompared.Basedonallsubjects，theleastabsolute shrinkageandselectionoperator（LASSO）regressionwasusedto identifyriskfactorsassociatedwithALNMburden，andthe synthetic minorityover-sampling technique（SMOTE）was applied toaddressclassimbalance.Five machine learning models includingsupportvectormachine，extremegadientbosting，randmforest（RF），K-nearesteighbors，ndonvolutioalneural network wereconstructed basedonthe screenedvariables.Receiver operating characteristic（ROC）curves and learning curves wereusedtoselecttheoptimalmodel，anditsgeneralizationabiltyacrossdiferentclinicalstagesasvaliatedintetetset. ResultsThere were significant diferences intumorTstage，maximum tumordiameter，time-intensitycurve（TIC）type，and

Ki-67expression between the trainingandtesting sets（all P <0.05），whileno significantdifferenceswereobserved intherest clinicaldataand imagingfeatures.Sixvariableswere selectedasassociatedwithALNMburdenbyLASSOandSMOTE：maximum tumordiameter，pathological type，molecularsubtype，tumorlocation，minimumapparent difusioncoeficientandTICtype，nd fivemachine learning modelswere constructed.Basedonthe diagnosticperformanceand learning curves，the RFmodel demonstratedcelntcallate，alancdperformane，ndoialstabilityonvergenceseed，ndgeeralzatiobility making ittheoptimal model.ROCcurve analysisshowedthattheRFmodelachievedtheareaunderthecurveofO.86（95%CI： 0.84-0.88），0.80（95%CI：0.75-0.85），and 0.83（95%CI：0.77-0.88）for diagnosing ALNMburdenin cT1-2N0M0，cT1NOM0， andcT2NoMo breastcancer patients，respectively.ConclusionTheRFmodel basedonclinical and imaging features performs well indiagnosingALNMburdeninbreastcancerpatientswithimaging-negativeaxilarylymphnodes，aidinginthe identification of low ALNM burden patients and demonstrating certain clinical value.

KEYWORDS Ultrasonography；MRI； Breast cancer；Axillary lymph node metastasis burden；Machine learning;'redictive model

根据2020年全球癌症统计报告1显示，女性乳腺癌的患病率高达 11.7% 。（剩余12240字）