术前MRI影像组学预测模型在非增强成人弥漫性胶质瘤分级诊断中的应用

打开文本图片集

lication of MRI radiomics predictive model in grading non-enhancing adult diffuse gliomas

LIUYanhongl，ANGYhnl2，HEZelong，WANGngxiao2，UJiayi3，ZHANGXuemeng2，LUJalin，NGul 1ChinesePLAMedicalScholBeijing853，hina;DepartmentfdiologyFirstMedicalCenterfineseGeel Hospital，Beijingo5，hina；DepartmentofRadiologyForthMedicalCenterofhinesePLAGeneralHospital，eijing 100048，China;eparmntofurosugeryrstdicalnterofnseealspialeiinga [Abstract]Objective：Toevaluate thevalueof preoperative MRIradiomicsmodelsingradingnon-enhancingadultdifuse gliomas.Methods：Aretrospectiveanalysisof153patients（16Olesions）withadultdifusegliomaswasconstructed.Al lesionshadnoenhancementonTWIcontrast-enhancement（TWI-CE）image.Amongthe16Olesions，103lesionswere clasifiedintothelow-gradegliomagroupand57lesionsinthehigh-gradegliomagroup，thenrandomlydivided into the trainingset（128lesions）andthevalidationset（32lesions）ataratioof8：2.TWIandADCimageswereco-registered with T 1 WI-CEimages，andROIsweredelineatedonTWIiamges.Atotalof1132featureswereextractedformTWI， TIWI-CEandADCimages，comprisingshape，first-orderhistogram，second-ordertextureandwaveletfeatures.Thefeatureswith stability（ICC>085）wereretained，andtheoptimalfeaturesubsetwasselectedviaelasticnetworkregression（ENR）comined withrecursivefeaturelimination（RFE）.Sixmachineleaningclasifiersereadoptedtoconstructthemodel，includinglogistic regresion，upportctorahine（adialsisuconerel，arel），K-eresteighor（K）isioee，ive Bayes.The predictiveperformanceofthemodelswasevaluatedusingtheAUC，accuracysensivity，specificityandFl-score. Results：Radiomicsmodelsefectivelypredictedhigh-gradeglioma.Themultiparametricmodel（TWI+TWI-CE+ADC）with KNNclasifierachievedAUCsof0.892and0.805，accuraciesof0805and0.719，sensitivitiesof0.915ad0.800，speciicties of 0.753and0.727，F1-scoresof 0.782and 0.506in the training set and thevalidationset.Conclusion ： Preoperative MRI radiomicsmodelsenableaccurategradingofnon-enhancingadultdifusegliomas，oferingcriticalsupportforclinical decision-making and patient management.

[Key words] Difuse glioma;High-grade glioma;Magnetic resonance imaging;Radiomics;Machine learning

脑胶质瘤是中枢神经系统最常见的原发性肿瘤，约占原发恶性脑肿瘤的 80%[1] 。（剩余9317字）