TABLE V.

COMPARSION BETWEEN THE PROPOSED 3D-GLCM MODEL WITH OTHER CNN BASED STRATAGIES. THE EVALUATION RESULTS ARE TWO-FOLD CROSS-VALIDATION METHOD.

Methods	Model	AUC	ACC	SEN	SPE	p-value
CT images based CNN	R-IMG	0.60±0.11	0.60	0.66	0.54	<0.0001
	C-IMG	0.67±0.07	0.64	0.69	0.59	<0.0001
	ResNet-18	0.63±0.06	0.57	0.68	0.46	<0.0001
	R-Hybrid	0.67±0.08	0.64	0.56	0.73	<0.0001
	C-Hybrid	0.70±0.07	0.66	0.73	0.59	<0.0001
	M-IMG (20)	0.81±0.06	0.74	0.87	0.60	<0.0001
	M-Hybrid (20)	0.83±0.06	0.76	0.76	0.76	<0.0001
	M-IMG (80)	0.68±0.10	0.63	0.76	0.50	<0.0001
	M-Hybrid (80)	0.73±0.06	0.68	0.79	0.55	<0.0001
	M-IMG (vote)	0.73±0.07	0.68	0.83	0.52	<0.0001
	3D-IMG (20)	0.84±0.05	0.77	0.82	0.72	<0.0001
	3D-IMG (80)	0.80±0.06	0.77	0.69	0.87	<0.0001
GLCM based CNN	R-GLCM	0.68±0.08	0.69	0.63	0.75	<0.0001
	C-GLCM	0.79±0.07	0.72	0.76	0.68	<0.0001
	M-GLCM	0.85±0.06	0.77	0.78	0.77	<0.0001
	3D-GLCM	0.91±0.05	0.87	0.90	0.71	1.0000

ACC, SEN and SPE are short for accuracy, sensitivity and specificity.