A novel robotic-based method to quantify and grade corneal neovascularization in live mice

Abstract

Purpose : Corneal neovascularization (CNV) is a common vision impairing condition affecting millions of people worldwide. At present, level of CNV representative of human clinical condition in mouse model is mostly performed manually with large variability. This study aimed to define a robotic-based method recapitutating clinical CNV grading by quantifying corneal neovascularization from corneal images taken from live mice with an automated computerized algorithm.

Methods : C57BL/6 mice (n=90) were employed to standarized method for reproducible quantification of CNV in live animals. CNV was produced using a topical alkali (10µl of 0.5M sodium hydroxide) application on the central cornea for 30 seconds. The initiation of blood vessels and progression of CNV were recorded for 5 weeks by collecting corneal images with stereomicroscope. An algorithm was developed to quantify density of vasculature (D), the existence of vasculature (E), length of vasculature (L), thickness of vasculature (T), and area of vasculature (A) in cornea. Over 1,000 stereomicroscopy images were used to train the model and develop algorithm. Animals were euthanized humanely and collected eye tissues were used for histology to verify levels of CNV grading performed by a computerized system.

Results : An automated computerized method (DELTA) quantifies density, existence, length, thickness and area of vasculature in corneal tissue. The algorithm of the DELTA method now made available to public at https://github.com/CIVA-Lab/CNV showed a high level of sensitivity, precision, specificity, dice, and accuracy of >90% in training and testing with an R-square value of 0.8234 in challenges. The developed DELTA method assigns Grade-0 to no CNV; Grade-1 to mild CNV; Grade-2 to moderate CNV; and Grade-4 to severe CNV with the images collected from the stereo microscope. The H&E staining and flat mount lectin immunofluorescences staining data well corroborated with the masked clinical biomicroscopic evaluation with slitlamp.

Conclusions : A newly developed robotic-based DELTA method for CNV grading in live mice is a convenient, reliable, and reproducible. It is expected to reduce variability and bias of manual methods.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.