Comparison of manual and automated endothelial cell density analysis in normal eyes and DSEK eyes

To compare automated endothelial cell density analysis with manual cell detection methods with 3 imaging devices.

In this prospective study, the corneal endothelium of 54 Descemet stripping endothelial keratoplasty (DSEK) eyes and 28 normal eyes was analyzed with a Nidek Confoscan 4 confocal microscope using a 20× noncontact lens and with Tomey EM-3000 and Konan Noncon Robo SP-8800 specular microscopes. Testing order was randomized. The Confoscan and Robo images were presented in a blinded fashion to an experienced technician for manual cell identification and analysis using the manufacturer's software. A different operator determined endothelial cell density using fully automated software associated with each imaging device. Agreement between methods was assessed by repeated-measures analysis of variance and post hoc Tukey analysis.

Manual cell identification on Robo and Confoscan 4 images produced comparable cell density measurements in normal eyes (P = 0.73) and DSEK eyes (P = 0.11). The Confoscan automated cell detection software differed significantly from manual cell detection in both normal and DSEK eyes (P = 0.0003 and P < 0.0001, respectively). The Robo automated cell detection software produced results comparable with manual cell detection in normal eyes (P = 0.082) but significantly overestimated cell density in DSEK eyes (P < 0.0001). The EM-3000 automated cell detection produced results comparable with manual cell detection in normal eyes (P = 0.067) and DSEK eyes (P = 0.49).

Only 1 of 3 automated cell detection programs produced cell density readings comparable with those obtained with manual cell identification; the other 2 automated programs significantly overstated endothelial cell density in DSEK eyes.