BMC Bioinformatics

Table 4 The FNR and the other metrics of the CNNs on the NINA and NIA datasets are reported. The image improvement techniques are active

From: Refactoring and performance analysis of the main CNN architectures: using false negative rate minimization to solve the clinical images melanoma detection problem

IIQ not active
Net	Data augmentation	SN	SP	PPV	FDR	FNR	FPR
AlexNet	None	0.87	0.90	0.86	0.15	0.13	0.10
AlexNet	Yes	0.84	0.91	0.87	0.14	0.16	0.09
DenseNet	None	0.56	0.82	0.77	0.23	0.29	0.18
DenseNet	Yes	0.64	0.74	0.56	0.44	0.19	0.26
Google InceptionV3	None	0.73	0.76	0.62	0.38	0.27	0.24
Google InceptionV3	Yes	0.39	0.60	0.29	0.71	0.57	0.40
GoogleNet	None	0.79	0.82	0.72	0.28	0.21	0.18
GoogleNet	Yes	0.45	0.63	0.48	0.52	0.54	0.37
MobileNet	None	0.81	0.72	0.45	0.55	0.14	0.28
MobileNet	Yes	0.32	0.61	0.29	0.71	0.37	0.38
ShuffleNet	None	0.61	0.74	0.60	0.40	0.35	0.26
ShuffleNet	Yes	0.36	0.60	0.45	0.55	0.52	0.39
SqueezeNet	None	0.23	0.43	0.43	0.57	0.22	0.27
SqueezeNet	Yes	0.43	0.62	0.41	0.59	0.41	0.37
VGG	None	0.58	0.83	0.76	0.24	0.27	0.17
VGG	Yes	0.82	0.59	0.40	0.60	*0.07*	0.24

Back to article page

ISSN: 1471-2105

Contact us

General enquiries: journalsubmissions@springernature.com