Machine learning analyses can differentiate meningioma grade by features on magnetic resonance imaging

摘要: OBJECTIVEPrognostication and surgical planning for WHO grade I versus II meningioma requires thoughtful decision-making based on radiographic evidence, among other factors. Although conventional statistical models such as logistic regression are useful, machine learning (ML) algorithms often more predictive, have higher discriminative ability, can learn from new data. The authors used an array of ML to predict atypical radiologist-interpreted preoperative MRI findings. goal this study was compare the performance standard methods when predicting grade.METHODSThe cohort included patients aged 18-65 years with (n = 94) 34) in whom obtained between 1998 2010. A board-certified neuroradiologist, blinded histological grade, interpreted all MR images tumor volume, degree peritumoral edema, presence necrosis, location, a draining vein, patient sex. trained validated several binary classifiers: k-nearest neighbors models, support vector machines, naive Bayes classifiers, artificial neural networks well grade. area under curve-receiver operating characteristic curve comparison across within model classes. All analyses were performed MATLAB using MacBook Pro.RESULTSThe 6 imaging demographic variables: sex, vein construct models. outperformed true-positive false-positive (receiver characteristic) space (area 0.8895).CONCLUSIONSML powerful computational tools that great accuracy.