Deep Learning in Toxicologic Pathology: A New Approach to Evaluate Rodent Retinal Atrophy.

Area of Science:

• Ophthalmology
• Toxicologic Pathology
• Computational Biology

Background:

• Manual quantification of retinal atrophy is labor-intensive.
• Retinal atrophy can be induced by therapeutics or light exposure.
• Accurate assessment is crucial for preclinical safety and efficacy studies.

Purpose of the Study:

• To explore deep learning (DL) for automating retinal atrophy assessment in rodents.
• To develop and validate a hybrid approach combining image processing and DL.
• To quantify atrophy in the outer and inner nuclear layers of the retina.

Main Methods:

• Utilized a DL approach based on the VGG16 model architecture.
• Trained, tested, and validated models using 10,746 image patches from rodent retinal whole slide images (WSIs).
• Combined conventional image processing with DL for quantification, validated by pathologist annotations.

Main Results:

• The DL approach efficiently quantified rodent retinal atrophy, particularly in the outer retina.
• Accuracy was validated against pathologist-annotated WSIs.
• The method successfully quantified the thickness of the outer and inner nuclear layers.

Conclusions:

• Deep learning significantly facilitates the evaluation of therapeutic and/or light-induced retinal atrophy in rodents.
• This study provides a template for training and validating DL models in preclinical toxicology.
• The developed method offers an efficient alternative to manual measurements for assessing retinal changes.