Abstract
Purpose :
This study aimed to develop an automated method for analyzing large series of retinal pigment epithelium (RPE) cells in flat mounts. Bulk analysis of such data sets helps us to measure individual cell morphometry, and to determine texture and spatial point properties of RPE sheets that reflect cell-cell and cell-environment interactions and cell organization in normal and diseased eyes.
Methods :
RPE flat mounts were compared from two groups of mice, rd10 and C57BL/6J at different ages (P30, P45, P60 and P100). The RPE sheet was prepared and imaged as before (Boatright et al. Mol Vis 2015; 21:40-60). Images were processed using MATLAB scripts and an automated “Cutbox selector”. CellProfiler software was used to perform morphometric analysis of cell shape, area, number of neighboring cells, and 20 other metrics.
Results :
A major bottleneck in processing images is to select parts of the RPE sheet that lacked dissection, processing, and staining artifacts. Manual selection was time-consuming. We implemented an automated method based on statistical analysis using k-means clustering of the tissue patterns to automatically generate and select artifact-free boxes from the RPE sheet. Further, the scripts numbered the boxes, and collected x,y coordinates automatically, and organized them for input into CellProfiler version 2.1.1, which then performed downstream analysis automatically. While the manual selection method took 1.5-2.5 hours for each eye, our automated counterpart reduced completion time to a few seconds. With this method, coverage of the whole RPE sheet was increased from about 20% to 50%. This method allows almost completely automated processing once images were collected, ensuring collection of more quantitative and unbiased data.
Conclusions :
Manual analysis of image files, such as images of retinal pigment epithelium (RPE) sheet, is difficult and in many cases, quantitative manual analysis of cell shapes is impossible. The risk of human errors, heavy time commitment, and sampling bias are just some of the drawbacks of manual implementations. This automated method proved useful for the analysis of RPE cells in flat mounts with greater accuracy, more efficient sampling of data points, and less human bias and errors. Our automated tool has become an integral part of our RPE sheet and cell analysis system. This approach would work well with many other histology or EM image sets.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.