Validation And Refinement Of An Automated Technique Of Counting Apoptosing Retinal Cells Imaged With DARC

Abstract

Purpose: : Apoptosing retinal cells imaged with the DARC (Detection of Apoptosing Retinal Cells) technology are normally counted manually. A novel automated technique of counting apoptosing retinal cells was recently developed using an algorithm developed on Matlab software. This was piloted with a single operator with very encouraging results. The method used for classification of images before Matlab analysis was subjective, and could not be used by inexperienced operators. The purpose of this study was to examine the validity and reliability of a new classification protocol for automated cell quantification.

Methods: : 88 rat eyes exposed to various neurotoxic insults and imaged with DARC were analysed by 3 inexperienced and masked operators. A new protocol was used in which each operator carried out the automated analysis using one of two cell size cut-off parameters. Matlab produced two cell-labelled images, of which the better labelled image was chosen by the operator through quick visual assessment. Apoptosing retinal cells in the same eyes were then counted manually by the 3 blinded operators. To test the consistency and reliability of the new method, Pearson’s R, Intraclass correlation coefficient (ICC) and Cronbach’s Alpha Reliability Coefficient were used.

Results: : The inexperienced operators were able to analyse the images with the refined protocol without difficulty. Analysis of the refined protocol revealed a statistically significant correlation between the mean manual and the mean automated cell counts: Pearson’s R (0.960, p<0.0001); ICC (0.977, p<0.0001) and alpha (0.977). Similar results were found between the individual operators: Pearson’s R (0.975, 0.922, 0.918 between paired manual/automated measurements; p<0.0001 in all cases); ICC (0.974, 0.912, 0.916 and p<0.0001) and alpha (0.987, 0.954 and 0.956). It was found that there was no statistically significant difference between the 3 operators: ANOVA (F 2.50, p<0.081) and the 3 automated measurements (F 0.473, p<0.624). Finally, the maximum time taken for automated analysis was one minute, in contrast to manual analysis which took 5-10 minutes.

Conclusions: : The automated technique of counting apoptosing retinal cells has been refined and validated to ensure that it is user-friendly, consistent and reliable. The time taken for automated analysis is shorter than the time taken for manual analysis, without affecting the accuracy of the results. This will enable utilisation of this technique for mass-scale image analysis. It is anticipated that this novel technique will be even faster when DARC is used in humans, as the larger size of apoptosing retinal cells will eliminate the need for setting two different Matlab cell-size parameters.

Keywords: imaging/image analysis: non-clinical • image processing • apoptosis/cell death