Quantification of retinal blood flow in swept-source Doppler OCT

Abstract

 

Purpose

 

Doppler OCT is an extension of standard OCT and is sensitive to moving particles due to signal decorrelation between consecutive A-lines, resulting in phase differences in the OCT signal. Based on Bayesian statistics the flow velocity is estimated quantitatively from these phase differences.

 

Methods

 

A high-speed fiber-based Doppler OCT system (Braaf et al., Opt. Express 2011) was used to acquire repeated A-lines with a time interval of 2.5ms using a backstitched B-scan pattern that measures each lateral tissue location twice (Braaf et al., Opt. Express 2012). Previously, Bouwens et al. (Opt. Let. 2014) used a model to directly calculate flow velocities from the mean and standard deviation of phase differences. Here, this model is extended to include influences of noise and a maximum a posteriori (MAP) estimator is introduced to determine flow velocities without the bias that occurs in the direct calculation.

 

Results

 

Figure 1 presents an en face plane from an age related macular degeneration (AMD) patient. It shows the removal of bias for flow estimation (grayscale is normalized to the highest flow). a) was processed with the method of direct calculation and b) with the MAP estimator. The backstitched scan pattern can be seen in the bias due to gradual changes in the background noise caused by revisitation error (i.e. the error in rescanning the same location). This induced an offset in method a) but the MAP estimator could isolate the flow from it. Figure 2 shows quantitative flow in the same en face plane processed with the MAP estimator to provide flow velocities. The determined flow velocities are within the expected range for retinal capillaries of 0.2 to 3.3 mm/s.

 

Conclusions

 

A new estimator was developed to quantify flow velocities of retinal capillaries directly from in vivo Doppler-OCT measurements. The method removes erroneous flow estimates due to revisitation errors and produces results that are in the expected velocity range. This method is potentially interesting for studying retinal pathology such as AMD.  

 

Flow estimation by a) direct flow calculation and (b) MAP estimator, showing removal of bias in the flow estimates due to backstitched scanning which appear as vertical stripes in the background signal. Black horizontal lines are motion artifacts which were taken out.

 

 

Three-pixel-layer en face plane of quantitative flow estimation with MAP. Flow velocity is displayed in false color.