Fig. 1

Proposed method for estimating venous pulse wave velocity. (A) Acquisition setup showing the patient seated at the video ophthalmodynamometer with the pulse oximeter affixed to their left index finger. (B) Acquired video frames over 3 cardiac cycles. (C) Extracted green channel images (shown in greyscale) after replacement of blurry video frames and alignment to the sharpest frame. (D) Manual segmentation/delineation of the optic disc boundary and the vessel of interest. The centerline is then given by the skeleton (medial axis) of the vessel mask. (E) Negative log transform of the pixel intensities in (C) so that intensity is related to blood column thickness and hence vessel diameter (Beer-Lambert law). (F) The harmonic regression model is fitted to each pixel time series in (E). The fitted model, first harmonic component, second harmonic component, and non-periodic component are shown for one of the centerline pixels. The first harmonic of the periodic component is shown for 3 centerline pixels. (G) Phase of the first harmonic and the amplitude of the periodic component both plotted against distance along the vessel centerline (from the point closest to the optic disc centre to the periphery of the optic disc). (H) Computation of PWV from the slope of the least squares line fitted to the distance-phase points.