Automated chest screening based on a hybrid model of transfer learning and convolutional sparse denoising autoencoder

Table 6 Comparison of accuracy, recall, F1 score, and AUC of the methods on test data by the deep network based on four data augmentation methods

Loss function	Data augmentation	Precision	Recall	F1	AUC
Cross entropy	Without augmentation	0.76	0.72	0.74	0.821
	Without augmentation	0.91	0.93	0.92	0.821
	SMOTE	0.58	0.82	0.68	0.813
	SMOTE	0.93	0.81	0.87	0.813
	Positive augmentation	0.67	0.75	0.81	0.815
	Positive augmentation	0.92	0.88	0.90	0.815
	4× augmentation	0.69	0.77	0.72	0.827
	4× augmentation	0.92	0.89	0.91	0.827
Focal loss	Without augmentation	0.76	0.73	0.74	0.828
	Without augmentation	0.91	0.93	0.92	0.828
	SMOTE	0.52	0.82	0.64	0.79
	SMOTE	0.93	0.76	0.84	0.79
	Positive augmentation	0.74	0.72	0.73	0.817
	Positive augmentation	0.91	0.92	0.91	0.817
	4× augmentation	0.7	0.74	0.72	0.821
	4× augmentation	0.9	0.9	0.91	0.821

ISSN: 1475-925X