Visualization of fundus vessel pulsation using principal component analysis

Abstract

Purpose: Spontaneous venous pulsation is one of the clinical signs with which to rule out elevated intracranial pressure and papilledema. More subtle pulsatile retinal movements are difficult to observe because of eye movements. Recording a fundus movie and aligning (registering) the images helps, but the images still contain distracting microsaccadic distortions and noise. The authors hypothesized that addressing these latter points should allow observation of minute pulsating features in fundus movies.

Methods: Principal component analysis (PCA), a basic form of blind source analysis, is applied to recorded fundus image sequences. The authors demonstrate this method in 5-second image sequences acquired with a near-infrared SLO (HRA+OCT Spectralis). The images are first registered to correct for eye drift, then microsaccade-distorted images are rejected, and the remaining image sequence is decomposed into principal components. Finally, a movie is constructed using the first five principal components (these had pulsatile features).

Results: Each of the processing steps (registration, cleaning, PCA filtering) improves the detection of pulsatile features, ultimately allowing clear visualization of spontaneous venous pulsation. Depending on the subject, additional features can be observed: pulsation amplitude of the arterial tree of approximately 10 μm, pulsation of arterioles down to 70-μm diameter, complete venous collapse, overall optic nerve head tissue pulsation, and mechanical links between veins and arteries.

Conclusions: By disentangling pulsatile motion from other dynamic components of retinal images, unprecedented resolution in physiologic motion of retinal vessel structure is achievable.