Intelligent Quantification of Retinal Ganglion Cells in the Entire Mouse Retina Based on Improved YOLOv5

Abstract

Purpose : To develop an automatic method to count specifically-labeled mouse retinal ganglion cells (RGCs) with improved accuracy and efficiency. This method can be used to assess the status of RGC degeneration in glaucoma mouse models.

Methods : An automated algorithm was developed based on improved YOLOv5. Five channels instead of a single channel with the Squeeze-and-Excitation block added was used to improve the accuracy of counting. The complete number of RGCs in an intact mouse retina is obtained by dividing the retina into small overlapping areas and counting followed by merging all divided areas together using the Non-Maximum Suppress algorithm. The degeneration of RGCs was induced by intravitreal injection of NMDA. RGCs were labeled with an anti-BRN3A antibody.

Results : The automated quantification result shows a very strong correlation (mean Pearson correlation coefficient approximately equal to 0.993) with that of manual counting for both normal mice and a glaucoma mouse model. Our model achieves a mean average precision of 0.981. Furthermore, the GPU calculation time for each mouse retina is less than one minute.

Conclusions : We have developed a new automatic RGC counting technique with great accuracy and speed. It will provide a convenient tool for researchers who are engaged in glaucoma research using mouse models to understand the pathogenesis of glaucoma and develop potential therapeutic drugs.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.