Table 3.

Meta-analysis results of sensitivity and specificity of machine learning in predicting lymph node metastasis in EC patients.

Variables	Model type	Training set			Validation set
		n	Sen(95%CI)	Spe(95%CI)	n	Sen(95%CI)	Spe(95%CI)
Radiomics Features
	Logistic regression	10	0.83(0.75~0.88)	0.79(0.76~0.82)	7	0.77(0.60~0.88)	0.86(0.75~0.93)
	Artificial neural network	2	0.77~0.86	0.66~0.94	1	0.89	0.75
	Support vector machine	2	0.75~0.81	0.75~0.87	1	0.71	0.72
	Ridge regression	1	0.7	0.86
	HoeffdingTree	1	0.81	0.87
	Convolutional neural network	2	0.80~0.83	0.90~0.91
	Overall	18	0.82(0.79~0.85)	0.83(0.79~0.87)	9	0.77(0.64~0.87)	0.84(0.74~0.91)
Radiomics+Clinical Features
	Logistic regression	7	0.90(0.84~0.94)	0.80(0.72~0.86)	6	0.78(0.62~0.88)	0.87(0.78~0.93)
	Artificial neural network	1	0.92	0.84	2	0.85~0.89	0.75~0.83
	Ridge regression	1	0.71	0.73
	Convolutional neural network	1	0.83	0.91
	Overall	10	0.88(0.84~0.92)	0.81(0.75~0.86)	6	0.81(0.70~0.89)	0.84(0.76~0.89)
Clinical Features
	Logistic regression	31	0.81(0.78~0.84)	0.75(0.71~0.79)	13	0.74(0.66~0.80)	0.79(0.75~0.82)
	Random forest	1	0.67	0.78	1	0.48	0.87
	Bayesian network				2	0.87~0.94	0.68~0.70
	Overall	32	0.81(0.77~0.84)	0.75(0.71~0.79)	16	0.75(0.67~0.82)	0.78(0.74~0.82)
Mayo					7	0.81(0.66~0.90)	0.59(0.38~0.77)