Table 2.

Characteristics of FAF images used in deep learning training.

Subject IDa	Fleck lesion type	RE		LE
		Fleck count	Fleck area (mm2)	Fleck count	Fleck area (mm2)	Dice scoreb
1c	Discrete	249	5.49	171	2.74
1c	Discrete	n/a		152	2.02
1c	Discrete	n/a		144	2.41
2	Diffuse	128	0.70	149	1.39
3a	Discrete	n/a		10	0.29
3b	Discrete	167	4.85	207	4.85	0.77
4a	Discrete	53	0.69	59	0.79	0.72
4b	Discrete	159	4.36	233	3.71
5	Discrete	78	2.59	57	2.44
6	Discrete	202	2.11	249	2.76
7a	Diffuse	202	2.04	182	1.34
7b	Diffuse	109	1.38	147	1.29	0.61
8a	Diffuse	5	0.05	7	0.07
8b	Diffuse	71	1.48	50	1.03	0.63
9a	Diffuse	37	1.03	109	1.57
9b	Diffuse	47	0.35	102	0.68	0.60
10	Discrete	145	4.29	154	3.74	0.80
11	Discrete	101	2.02	124	2.17	0.70
12	Discrete	48	0.56	29	0.44	0.69
13	Discrete	34	0.73	35	0.79	0.57
14	Diffuse	57	0.43	67	0.48	0.33
15	Discrete	29	0.50	10	0.86
16	Discrete	50	2.85	89	2.55
17c	Discrete	n/a		64	0.71
17c	Discrete	n/a		52	0.60
18	Discrete	n/a		55	2.43
19	Discrete	n/a		155	5.55

n/a not applicable.

^aIndividuals from the same family are indicated by a lower case letter following the same number; all were siblings except for subjects 3a and 3b, who were mother and son.

^bDice score between manual and deep learning segmentation, only available in 10 left eye images used for validation.

^cLongitudinal serial FAF images were used as part of the training set in subjects 1 and 17.