. 2021 Jul 16;11:14590. doi: 10.1038/s41598-021-93905-2

Table 2.

Literature overview of eye and tumor segmentation methods and performances.

Reference	Model used	Data set	Pulse sequence	Performance (DSC)
Reference	Model used	Data set	Pulse sequence	Sclera	VH	Lens	Tumor
Current	MV-CNN	MR, 29 RB 17 healthy eyes	T1c, T1, T2, FIESTA	0.84	0.93	0.91	0.84
De Graaf 2019¹⁴	ASM + 3D U-Net	MR, 24 RB, 11 healthy eyes	T2	0.90*	–	0.81	0.65
Nguyen 2019¹⁶	2D U-Net + ASM / CRF	MR, 24 UM	T1, T2	–	–	–	0.84
Nguyen 2018¹⁵	3D ASM	MR, 7 UM, 30 healthy eyes	T1	0.95*	0.92	0.88	-
Nguyen 2018¹¹	3D U-Net	MR, 32 RB eyes, 40 healthy eyes⁺, multi-center	T1, T2	0.95*	–	0.87	0.59
Ciller 2017¹⁰	3D ASM + 3D CNN	MR, 16 RB eyes	3D T1c, T1, T2	0.95*	0.95	0.86	0.62
Ciller 2015¹²	3D ASM	MR, 24 healthy eyes	3D T1c	0.95*	0.95	0.85	–
Beenakker 2015³²	Topo-graphic map	MR, 17 healthy eyes	3D IR TGE	No DSC reported
Rüegsegger 2012¹³	3D ASM	CT, 17 healthy eyes	Does not apply	0.95*	–	0.91	–
Bach Cuadra 2010²¹	3D parametric model	US, CT, 3 RB eyes	Does not apply	0.91*	-	0.77	-

DSC dice similarity coefficient; VH vitreous humour; MV multi-view; CNN convolutional neural networks; MR magnetic resonance; ASM active shape model; CRF conditional random field; UM uveal melanoma; RB retinoblastoma; IR inversion recovery; TGE turbo gradient echo; CT computed tomography; US ultrasound; T1c T1 with gadolinium contrast.

*Includes vitreous humour.

⁺Includes child and adult scans.