Stratégies de comptage par analyse d'image de l'endothélium des cornées conservées en organoculture : détection de contour versus pointage du centre . Counting strategies for endothelial assessment of organ cultured corneas using image analysis: comparison of border versus center method
Abstract
Aim To compare two different cell count strategies, border and center methods for endothelial assessment during corneal storage. Material and Methods Seven observers determined the endothelial cell density (ECD), coefficient of variation (CV) of cell area, and percentage hexagonality of 30 organ cultured corneas by the border (contour detection and manual retouch) and center method (indicating cell centers) using Sambacornea analyser. Interobserver variability for ECD and agreement between the two methods was determined. The accuracy of the center method was verified by counting on ten standard photolithographic mosaics (called keratotest) by border (reference method) and center methods, first as fast and next as precise as possible (fast and slow modes respectively) and the time noted. Data between border and center methods was compared using nonparametric paired tests. Results For stored corneas, interobserver variability was ±9.6% (95%CI [6.5-12.7]) for border method and ±9.3% (95%CI[6.3-12.3]) for center method. ECD [mean±SD (range), median] was [2948±565 (1644-3878), 3081] and [2961±568 (1736-3914), 3037] respectively and showed excellent correlation (Pearson coefficient r=0.998, p<0.001). The center method underestimated the CV by a mean 9.5%95%CI [8.3-10.7] and overestimated the hexagonality by a mean 2.5% (95%CI [1.4-3.7])(p<0.001 for both). Precision of indication of cell center influenced only the CV (mean for slow and fast modes being 3.6±0.7% and 7.4±1.3% respectively, p<0.001) but was more time consuming (p=0.005). Discussion The center method gives accurate and reproducible measurement of ECD and is comparable to the border method (reference). However, morphometric evaluation is not fully reliable. Conclusion Given that center method is easier to use in poor quality images, we recommend its use only for these cases.