Table 3.

Comparison with the state-of-the-art works for the classification of lung cancer subtypes. This work used the results of the GBRT algorithm based on the augmented images of the training set, the original images of the test set, and used RS method to split for 5 times

	Works	Lung subtypes	Samples	Methods	No. of features	Image modal	Results
(1)	Wu W [17]	ADC, SCC	350	Naive Bayes	440	CT	AUC 0.720
(2)	Saad M [18]	ADC, SCC, LCC	317	SVM	624	CT	ACC 0.783 AUC 0.863
(3)	E L [19]	NSCLC, SCLC	278	SVM	1695	CT	AUC 0.741
(4)	Liu J [58]	ADC, SCC, LCC, NOS	349	SVM	1029	CT	ACC 0.860
(5)	Han Y [60]	ADC, ACC	1419	VGG16	No feature extraction	PET/CT	ACC 0.841 AUC 0.903
(6)	This work	SCC, LCC	169	GBRT	No feature extraction	CT	ACC 0.960 ± 0.005 AUC 0.984 ± 0.004

ADC  adenocarcinoma, SCC squamous cell carcinoma, LCC large-cell carcinoma, NOS not otherwise specified, NSCLC non-small cell lung cancer, SCLC small cell lung cancer, ACC accuracy, AUC area under the curve, SVM support vector machine