Related Experiment Video
Updated: Aug 6, 2026
12:28
Quantification of Vascular Parameters in Whole Mount Retinas of Mice with Non-Proliferative and Proliferative Retinopathies
Published on: March 12, 2022
Deep Learning-Guided Retinal Vascular Morphometric Quantification in Cerebral Autosomal Dominant Arteriopathy with
Hyungwoo Lee1,2, Na-Kyung Ryoo1, Said Arevalo-Alquichire1
1Schepens Eye Research Institute of Massachusetts Eye and Ear and Department of Ophthalmology, Harvard Medical School, Boston, Massachusetts.
Ophthalmology Science
|July 20, 2026
Summary
Explainable deep learning accurately identifies retinal vascular changes in mouse models of CADASIL. This method quantifies focal dilation and beading, aiding in the development of biomarkers for this condition.
Area of Science:
- Ophthalmology and neuroscience research.
- Medical imaging and computational pathology.
- Genetics and molecular biology.
Background:
- Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) is a genetic disorder affecting small blood vessels.
- Retinal imaging may offer insights into CADASIL pathology, but subtle changes are difficult to detect.
- Current methods for analyzing retinal vasculature lack the sensitivity to capture focal abnormalities.
Purpose of the Study:
- To develop and validate an explainable deep learning (DL) workflow.
- To localize and quantify focal retinal luminal pathology using fundus fluorescein angiography (FFA).
- To investigate NOTCH3 variant knock-in mouse models of CADASIL.
Main Methods:
- Trained VGG16 classifiers on FFA images from wild-type and NOTCH3 mutant mice.
- Utilized Grad-CAM++ and occlusion sensitivity maps to identify disease-relevant regions of interest (ROIs).
- Applied morphometry and a vessel beading index (VBI) for quantitative analysis of vessel diameter and periodic oscillations within ROIs.
Main Results:
- Whole-field analysis revealed generalized large-vessel dilation and reduced tortuosity in mutants.
- ROI-restricted analysis significantly amplified focal pathology, showing increased large-vessel maximum diameter and VBI in mutants.
- The developed DL framework demonstrated high classifier discrimination and robustness across analyses.
Conclusions:
- Explainable DL effectively localizes disease-informative retinal vessel segments on FFA.
- The workflow enables sensitive quantification of focal luminal dilation and beading, overcoming limitations of whole-field analysis.
- This approach holds potential for developing retinal biomarkers and monitoring CADASIL progression.
