A Deep Learning Approach to Measure Visual Function in Zebrafish.

Area of Science:

• Neuroscience and ophthalmology research
• Zebrafish as a model organism for visual behavior studies

Background:

• Optokinetic reflex (OKR) in zebrafish is crucial for modeling human eye diseases and drug testing.
• Conventional OKR analysis methods struggle with low-contrast or hypopigmented subjects.
• Existing techniques often require expensive software or are limited by binarization.

Purpose of the Study:

• To introduce a novel deep learning pipeline for robust optokinetic reflex (OKR) analysis in zebrafish.
• To demonstrate the pipeline's effectiveness in quantifying eye movements across varying contrast and pigmentation.
• To validate the deep learning approach against traditional methods.

Main Methods:

• Utilized a ResNet-50 model within the DeepLabCut framework for eye tracking.
• Applied the pipeline to wild-type and albino zebrafish larvae (5 days post-fertilization).
• Recorded and analyzed eye movements using both the deep learning pipeline and conventional software.

Main Results:

• The deep learning pipeline achieved accuracy comparable to traditional OKR analysis methods.
• The model successfully quantified eye movements in hypopigmented (albino) zebrafish larvae.
• Bland-Altman tests confirmed the reliability and consistency of the deep learning approach.

Conclusions:

• The developed deep learning pipeline offers a robust and accessible method for OKR analysis in zebrafish.
• This approach overcomes limitations of traditional methods, particularly for studies involving low-contrast or hypopigmented models.
• The pipeline is adaptable for broader behavioral studies in zebrafish, including those relevant to ocular and neurological diseases.