Integrating Retinal Segmentation Metrics with Machine Learning for Predictions from Mouse SD-OCT Scans.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Machine Learning in Biology

Background:

• Optical Coherence Tomography (OCT) is crucial for non-invasive retinal imaging.
• Retinal layer thickness metrics can serve as biomarkers for biological variations.
• Machine learning (ML) offers advanced analytical capabilities for complex biological data.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in predicting biological features (mouse age and species) from retinal segmentation metrics.
• To determine if ML can categorize mice based on retinal structural data derived from OCT scans.

Main Methods:

• Retinal layer thickness data from C57BL/6 and DBA/2J mice were segmented from SD-OCT scans.
• Twenty-two ML models were trained to predict mouse groups (species and age).
• A neural network model was optimized, and performance metrics (accuracy, AUC, sensitivity, specificity, precision, F-1 score) were evaluated.

Main Results:

• Neural networks demonstrated significantly higher validation accuracy compared to other ML classifiers (p=0.005).
• The optimized neural network achieved 92.31% accuracy for species classification (C57BL/6 vs. DBA/2J) with an AUC of 0.9762.
• The model also achieved 82.05% accuracy for age group differentiation with an AUC of 0.9064.

Conclusions:

• ML analysis of retinal segmentation metrics from OCT scans is effective for categorical predictions in mouse models.
• This method provides a valuable, non-invasive approach for experimental research.
• Future expansion with histopathological and functional data may enhance predictive power for complex clinical and experimental outcomes.