Table 2.

Methods used for segmentation of skin lesions.

Approach	Study	Method	Preprocessing	Dataset	Images
Supervised	[2]	Deep regional CNN and FCM clustering	Yes	ISIC 2016	1279
	[12]	Deep convolutional network	Yes	PH2	200
				ISBI 2017	2750
	[26]	Saliency based	Yes	ISIC 2017	2150
	[30]	FCN based	Yes	ISIC 2016	1279
				PH2	200
	[31]	Recurrent, residual convolutional neural network	Yes	ISIC 2017	2000
	[32]	CNN based ensemble	Yes	ISIC 2018	2594
	[33]	Hybrid learning, particle swarm optimization	Yes	ISIC 2017	550
	[34]	Semantic segmentation based on u-Net	Yes	ISIC 2018	2594
	[35]	R2AU-Net	No	ISIC 2018	2594
	[59]	Deep convolutional encoder-decoder	No	PH2	200
Unsupervised	[8]	Statistical region merging	Yes	Private	90
	[9]	Thresholding	Yes	ISIC 2017	600
	[10]	Stochastic region merging	No	PH2	200
				ISIC 2018 validation	100
				ISIC 2018 test	1000
	[17]	Thresholding	Yes	Private dataset	85
	[19]	Saliency and thresholding	Yes	PH2	18
				ISBI 2016	13
	[20]	K-means clustering	Yes	Dermatology information system andDermQuest	50
	[21]	K-means clustering	Yes	Atlas dermoscopy dataset	80
	[22]	Fuzzy C-Means clustering	Yes	UMCG	170
	[23]	Data clustering	Yes	PH2	200
				ISIC (2016–2019)	5400
	[24]	Saliency	No	EDRA	566
				PH2	200
				ISBI 2016	900
	[25]	Saliency	Yes	PH2	200
				ISBI 2016	900
	[27]	Saliency	Yes	PH2	50
				ISBI 2016	70
	[28]	Saliency and thresholding	Yes	PH2	200
				ISBI 2016	900
	[29]	Multi scale superpixel segmentation	No	PH2	200
				ISBI 2016	900
	[37]	Thresholding and edge detection	Yes	PH2	200
	[38]	Saliency	Yes	PH2	200
	[39]	Region merging	Yes	PH2	200
				ISIC 2017	900
	[40]	Thresholding	Yes	PH2 Mednode DermNet	992
	[50]	Thresholding and GraphCut	Yes	DSSA	294
	[52]	Partially homomorphic POB number system	Yes	PH2	200
				ISBI 2016	1279
				ISBI 2017	2600
	[60]	Superpixel clustering and thresholding	No	PH2	200
	Ours	Saliency-based color histogram clustering with thresholding	No	PH2	200
				ISIC 2018	2594
				HAM10000	10,015