SLOctolyzer

SLOctolyzer segments and quantifies retinal vessels from scanning laser ophthalmoscopy (SLO) images to extract reproducible and clinically meaningful retinal vascular parameters.

Key Features:

• Deep Learning Segmentation: Uses deep learning methods to delineate retinal anatomy including retinal vessels, fovea, and optic disc.
• Vascular Measurement Module: Quantifies complexity, density, tortuosity, and caliber of segmented retinal vessels.
• Performance on Internal Test Data: Achieves Dice scores of 0.91 for all-vessels, 0.84 for arteries, 0.85 for veins, 0.94 for optic disc, and 0.88 for fovea.
• Processing Speed: Processes a 768 x 768 pixel macula-centered SLO image in under 20 seconds and a disc-centered SLO image in under 30 seconds using a laptop CPU.
• Manual Correction and Re-computation: Allows manual correction of segmentations and re-computing of vascular metrics.

Scientific Applications:

• Retinal Vascular Assessment: Enables extraction of retinal vascular measurements from large OCT image sets for linking to ocular or systemic diseases.
• Pathology Evaluation: External validation against severe retinal pathology showed decreased performance with Dice scores of 0.72 for arteries, 0.75 for veins, and 0.90 for optic disc.
• Reproducibility Studies: Demonstrates good reproducibility with mean differences for fractal dimension of -0.001, density of -0.0003, caliber of -0.32 microns, and tortuosity density of 0.001.

Methodology:

SLOctolyzer includes two main modules: a segmentation module that uses deep learning methods to delineate retinal anatomy and detect the fovea and optic disc, and a measurement module that quantifies the complexity, density, tortuosity, and caliber of the segmented retinal vessels.