Longitudinal Analysis of Mouse SDOCT Volumes

Area of Science:

• Ophthalmology
• Biomedical Imaging
• Animal Models

Background:

• Spectral-domain optical coherence tomography (SDOCT) is increasingly used in animal studies for non-invasive, longitudinal research.
• Inconsistent scan orientation due to lack of anatomical landmarks hinders accurate longitudinal analysis in animal SDOCT studies.
• Accurate registration methods are crucial for reliable tracking of disease progression and drug efficacy in animal models.

Purpose of the Study:

• To develop and validate an automated method for registering mouse SDOCT volumes.
• To enable accurate longitudinal tracking of ocular changes in animal models using SDOCT.
• To overcome orientation inconsistencies in serial SDOCT scan acquisition.

Main Methods:

• Automated segmentation of retinal blood vessels and optic nerve head (ONH) using pixel classification.
• Iterative Closest Point (ICP) algorithm for registering follow-up vessel maps to baseline scans.
• Training a random forest classifier with 18 SDOCT volumes from a light damage model.
• Validation using leave-one-out analysis and qualitative assessment on a secondary dataset.

Main Results:

• High accuracy in segmenting retinal blood vessels (AUC=0.93) and optic nerve head (AUC=0.98) using the trained classifier.
• Successful registration of SDOCT volumes using the proposed framework, including retinal vasculature segmentation and ICP registration.
• Qualitative assessment confirmed no registration failures in a secondary set of scans from a light damage model.

Conclusions:

• The proposed automated registration method effectively addresses orientation inconsistencies in mouse SDOCT volumes.
• This framework enables reliable longitudinal tracking of ocular changes, crucial for disease mechanism and drug efficacy studies in animal models.
• The validated method enhances the utility of SDOCT in preclinical research, facilitating non-invasive, long-term monitoring.