Table 4.

Performance results of pre-trained CNN models based on ROI using GAC method for early diagnosis and differentiation of skin cancer and other skin lesions.

Models	Type of Lesion	AUC %	Sensitivity %	Accuracy %	Precision %	Specificity %
AlexNet	Scc	70.2	63.4	63.5	69	99.1
	Akiec	72.4	66.2	66.1	71	98.6
	Bcc	92.3	90.8	91.3	92.5	98.7
	Bkl	93.8	91.7	91.8	90.3	99.4
	Df	68.9	58.3	58.3	54.9	99.9
	Mel	92.5	91.2	91.5	92	97.7
	Nv	98.1	96.9	97.1	96.1	96.3
	Vasc	75.2	65.2	64.7	67.3	99.5
GoogLeNet	Scc	69.1	63.1	62.7	67.5	99.2
	Akiec	75.6	68.3	67	76.6	98.8
	Bcc	94.6	92.8	92.9	89.8	97.6
	Bkl	93.4	90.4	89.5	90.4	99.3
	Df	66.2	54.3	54.2	56.5	99.8
	Mel	94.2	92.7	92.6	90.2	97.5
	Nv	97.1	95.8	95.8	96.3	96.4
	Vasc	84.6	70.9	70.6	69.2	99.7
VGG16	Scc	62.2	53.4	53.2	54.9	99.3
	Akiec	67.1	62.3	62.1	69.7	98.5
	Bcc	92.3	89.8	89.6	88.8	97.8
	Bkl	94.2	91.4	90.7	88	98.6
	Df	50.2	44.2	43.8	60	99.5
	Mel	92.9	89.7	90	89.5	97.7
	Nv	97.5	95.6	95.8	95.7	96.2
	Vasc	98.6	70.2	68.6	62.5	99.7