文章题目
evelopmentandvalidationofanendoscopicimages-baseddeeplearningmodelfordetectionwithnasopharyngealmalignancies
研究人员:ChaofengLi
发表时间：Sep25
期刊名称：CancerCommun
影响因子：5.
分区：2区
01
核心亮点
TheeNPM-DMoutperformedexpertsindetectingnasopharyngealmalignancies.Moreover,thedevelopedmodelcouldalsoconductautomaticsegmentationofmalignantareafromtheconfusingbackgroundofnasopharyngealendoscopicimagesefficiently,showingpromisingprospectsinbiopsyguidancefornasopharyngealmalignancies（eNPM-DM在检测鼻咽恶性肿瘤方面表现优于专家。此外，该模型还可以有效地从鼻咽内窥镜图像混乱的背景中进行恶性区域的自动分割，在鼻咽恶性肿瘤活检指导中具有广阔的应用前景）.
02
思路与方法
1.回顾性收集接受鼻咽常规临床筛查的患者；
2.建立模型；
3.利用DICE等指标评价模型性能。
03
图文摘要
Abstract
Backgrounduetotheoccultanatomiclocationofthenasopharynxandfrequentpresenceofadenoidhyperplasia,thepositiverateformalignancyidentificationduringbiopsyislow,thusleadingtodelayedormisseddiagnosisfornasopharyngealmalignanciesuponinitialattempt.Here,weaimedtodevelopanartificialintelligencetooltodetectnasopharyngealmalignanciesunderendoscopicexaminationbasedondeeplearning.
Methods:Anendoscopicimages-basednasopharyngealmalignancydetectionmodel(eNPM-DM)consistingofafullyconvolutionalnetworkbasedontheinceptionarchitecturewasdevelopedandfine-tunedusingseparatetrainingandvalidationsetsforbothclassificationandsegmentation.Briefly,atotalof28,qualifiedimageswerecollected.Amongtheseimages,27,biopsy-provenimagesfromindividualsobtainedfromJanuary1st,,toDecember31st,,weresplitintothetraining,validationandtestsetsataratioof7:1:2usingsimplerandomization.Additionally,imagesobtainedfromJanuary1st,,toMarch31st,,wereusedasaprospectivetestsetto
