Abstract
Purpose :
Retinal manifestations of both ocular and systemic diseases, such as Alzheimer’s, have been well documented in the literature. Diagnostic applications using light, including hyperspectral imaging (HSI), can detect structural and functional changes based on light-tissue interaction. When coupled with texture analysis, this permits valuable information in differentiating healthy vs. disease. In this study, we hypothesize that different spatial-spectral features of the retina and choroid can be identified.
Methods :
HSI images of the retina are captured between 450-905 nm in increments of 5 nm under mydriatic conditions, resulting in 92 images obtained in ~1 second. This image cube is then normalized and registered to align key anatomical landmarks, then pixel texture features (how pixels vary relative to their neighbors in specific anatomical zones) are extracted.
Results :
To date over 2000 spatial-spectral features have been identified, each providing idiosyncratic information about the cellular and physiological properties of the retina; 2 are outlined here.
Feature 1 (Fig 1) assesses pixel distribution in retinal background, excluding vessels, based on texture analysis of images from 510 to 630 nm.
Feature 2 (Fig 2) assesses pixel distribution inside vessels based on texture analysis of images from 480 to 600 nm.
We note a unique pixel distribution for subject A vs. B, demonstrated in C, respectively.
Although the exact cause for pixel variations is unknown, possible contributors to these feature differences include but not limited to:
Amount, location and oxygen content of hemoglobin in vessels or tissue
Altered thickness of retinal layers (RNFL, IPL)
Protein, lipid, macular pigment, lipofuscin accumulation or pigment reorganization
These changes alter light absorption, scattering, and depth of penetration in tissue producing different pixel texture signatures that are wavelength dependent, and correlate with the phenotypic status of the retina.
Conclusions :
In this study HSI of the retina was combined with image processing using texture analysis to resolve spatial-spectral patterns of pixel intensity variability that result from structural and molecular organization of the retina. These patterns provide a retinal phenotype that could be employed as a multi-factorial assessment of tissue status, and in diagnosis of both retinal and systemic pathologies with manifestation in the eye.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.