Purpose
Increase in acellular capillary numbers is the pathologic hallmark of diabetic retinopathy. Traditionally, the acellular capillaries are enumerated by blinded investigators either directly through a microscope or via manual counting of captured images. However, this system is laborious, time consuming, and often shows inconsistency amongst researchers. The purpose of this study is to create a computer based algorithm that will assess the acellular capillaries of the retina, consistently reducing the human error and time.
Methods
The retinas of control and diabetic mice were processed using trypsin digestion and the high resolution .tiff images of retinal quadrants. We used the Python programming language using assorted open source package to write a custom-designed code. The images underwent a Gaussian blur and noise reduction to clean up the imperfections of the image. We used a purpose-built k-means clustering algorithm to group similar parts of the image together. We generated the paths in each image by converting all non-white elements to black. These images were then processed for Medial Access Transform (MAT) to create the skeleton as well as to find the distance from the skeleton to the edges formed in the above steps. The colors represent the distance from the edges, in which red is the largest distance and purple is the shortest distance. Then the locations where the skeleton is purple and connects to another color on both ends were counted. This count is the number of acellular capillaries.
Results
We have developed a precise algorithm with improved accuracy to enumerate the numbers of acellular capillaries. This algorithm can be used to quickly count the acelluar capillaries in diabetic retinas and to create a standard for retinopathy assessment. Moreover, this algorithm is compatible with open source image analysis programs, enabling ease of access to the users.
Conclusions
We have designed an automated computer-based system to enumerate the acellular capillaries in diabetic retina. This computer based automated system will enhance consistency in retinopathy assessment and reduce time for analysis.