Table 3.

Durations and failure percentages of the different analysis methods according to image quality (excluding image capture, which lasted about 1 minute)

	“Adequate” image			“Medium” image			“Good” image			All grades
Analysis method	Time	Failures	No	Time	Failures	No	Time	Failures	No	Time	Failures	No
Mono-image 50	174	0%	17	110	0%	79	79	0%	84	101	0%	180
	(98, 69–387)			(86, 43–600)			(27, 38–168)			(72, 38–600)
Tri-image 50	141	0%	11	124	0%	23	96	0%	26	115	0%	60
	(49, 95–260)			(49, 59–243)			(26, 62–141)			(44, 59–260)
Mono-image 300	698	59%	17	547	8%	79	356	0%	84	456	9%	180
	(287, 405–1230)			(234, 240–1200)			(126, 145–781)			(216, 145–1230)
Tri-image 300	577	0%	11	486	0%	23	367	0%	26	451	0%	60
	(147, 399–767)			(110, 258–689)			(140, 138–693)			(152, 138–767)

Mean time required was just under 2 minutes to count 50 cells and 7.5 minutes to count 300. With rare exceptions, automated analysis was far shorter than a manual count, which took 10–15 minutes. It should be noted that video image capture time allows the cornea to be returned very quickly to its storage medium. In the case of good images, analysis times were significantly shortened, falling to about 1.5 minutes for 50 cells (p<0.001) and to 6 minutes for 300 cells (p<0.001). Analysis time in tri-image mode was not longer than in mono-image mode, whichever the cell count. On the contrary, analysis time was substantially shorter for adequate images (not significant) because of multiple zone selection or a properly visible mosaic, which reduced the need for touch ups. Note finally that the only failed analyses were in mono-image 300 mode (p=0.014) and for medium or adequate images. Mean durations are expressed in seconds (SD, range).