A perspective on the diagnosis of cracked tooth: imaging modalities evolve to AI-based analysis

Table 3 Summary of convolutional neural network for crack segmentation

Methods	Models	Reference	Size of images	Precision (%)	Recall (%)	F1 (%)	Size of data sets
Image classification	DCNN	[111]	256 × 256	99.09			60,000
Image classification	DCNN	[110]	256 × 256	98			40,000
Object detection	YOLO	[118]	448 × 448	83.54	79.93		2000
	YOLO-v2	[21]	227 × 227	89			990
	YOLO-v2	[119]	416 × 416	88.51	87.1	87.8	9053
	YOLO-v3	[120]	416 × 416	89.16	91.16		1500
	YOLO-v3	[121]	480 × 600	88	91.16		4000
	Faster R-CNN	[115]	1865 × 2000	78.53	85.56		3000
		[116]		96.3			5966
		[117]	500 × 375	90.2			2366
Semantic segmentation	FCN (VGG19)	[126]	224 × 224	81.7	78.97	79.95	> 800
	FCN (VGG16)	[128]	227 × 227	90	78.97	89.3	> 800
	U-Net	[131]	572 × 572	92.46	82.82	87.38	118 (CrackForest)
		[136]	512 × 512	92.12	95.7	93.88	118 (CrackForest)
		[125]	48 × 48	90	91	90	57
		[134]	320 × 320	94.94	93.55	96.37	118 (CrackForest)
		[133]	256 × 256	97.02	94.32	95.55	118 (CrackForest)
		[137]	480 × 320	91.45	88.67	90.04	1200
		[138]		97.31	94.28	95.75	118 (CrackForest)
	CrackU-net	[132]	1024 × 1024	98.56	97.98	98.42	3000
	SegNet	[130]	360 × 480	90.92	97.47	79.16	1021
	SegNet	[129]	608 × 608	80.31	80.45	79.16	504
	CrackSeg	[139]	256 × 256	98	97.85	97.92	8198
	SDDNet	[140]	513 × 513	87.4	87	87.2	537
	FPCNet	[141]	288 × 288	97.48	96.39	96.93	118 (CrackForest)

ISSN: 1475-925X