Machine learning-based MRI radiomics for assessing the level of tumor infiltrating lymphocytes in oral tongue squamous cell carcinoma: a pilot study

Table 2 Comparison of the radiomics features selected from combined sequences between the two groups and diagnostic performance of the features

Code	Sequence	Image type	Feature class	Feature name	Low TILs	High TILs	P value	AUC (95% CI)
RF1	ceT1WI	LoG_{3 mm}	Histogram	Kurtosis	3.04 (2.79, 3.48)	2.84 (2.52, 3.05)	0.011	0.681 (0.554, 0.808)
RF2	ceT1WI	Wavelet_HH	Histogram	Skewness	0.11 (0.04, 0.34)	−0.02 (− 0.10, 0.06)	< 0.001	0.743 (0.621, 0.865)
RF3	ceT1WI	Original	GLCM	ClusterShade	−1032 (− 2032, − 146)	−110 (− 1154, 986)	0.035	0.650 (0.5317, 0.783)
RF4	ceT1WI	Original	GLDM	DependenceVariance	0.68 (0.50, 1)	0.56 (0.49, 0.66)	0.025	0.660 (0.530, 0.790)
RF5	ceT1WI	LoG_{1 mm}	Histogram	Minimum	−115 (− 143, − 101)	−104 (− 121, − 91)	0.016	0.672 (0.543, 0.801)
RF6	T2WI	Wavelet_HL	GLCM	JointEntropy	8.32 (8.02, 8.68)	8.69 (8.41, 8.99)	< 0.001	0.751 (0.633, 0.868)

AUC: area under the receiver operating characteristic curve; ceT1WI: contrast-enhanced T1-weighted imaging; GLCM: gray-level cooccurrence matrix; GLDM: gray-level dependence matrix; RF: radiomics feature; T2WI: T2-weighted imaging

ISSN: 1471-2342