Diagnostic Accuracy of Machine Learning-Based Radiomics in Grading Gliomas: Systematic Review and Meta-Analysis

Table 3

Summary of the results evaluated in the reviewed studies.

Study and year	Method	Algorithm	Dataset/HGG-LGG	MRI sequence	Best performance				Limitation
					AUC	DA (%)	Sen (%)	Spe (%)
Cho et al. 2018	Classic machine learning	Multiple algorithms	WHO II–IV (n = 285)/210-75	T1, T1-C, T2, T2- FLAIR	0.94	92.92	97.86	79.11	No dataset separation information for training and testing cohort. Sample imbalance size between LGG and HGG.
Tian et al. 2018	Classic machine learning	SVM	WHO II–IV gliomas (n = 153)/111-42	Multiparametric	0.99	96.80	96.40	97.30	Sample imbalance sample size between LGG and HGG.
Hashido et al. 2018	Classic machine learning	Logistic regression	WHO II–IV (n = 46)/31-15	ASL, PWI (DSC)	0.96	NA	89.30	92.90	Small sample size. Small sample size used in the training set. Large feature number than the total sample size.
Vamvakas et al. 2019	Classic machine learning	SVM	WHO I–IV (n = 40) 20-20	Multiparametric	0.96	95.50	95	96	Small sample size.
Zhao et al. 2020	Classic machine learning	RF	WHO II-III gliomas (n = 36) 17-19	T1-C, T2- FLAIR	0.86	78.10	78.30	77.80	Small sample size. Large feature number compared to the total sample size.