Quality improvement of OCT angiograms with elliptical directional filtering

We present a method of OCT angiography (OCTA) data filtering for noise suppression and improved visualization of the retinal vascular networks in <i>en face</i> projection images. In our approach, we use a set of filters applied in three orthogonal axes in the three-dimensional (3-D) data sets. Minimization of artifacts generated in B-scan-wise data processing is accomplished by filtering the cross-sections along the slow scanning axis. A-scans are de-noised by axial filtering. The core of the method is the application of directional filtering to the C-scans, i.e. one-pixel thick sections of the 3-D data set, perpendicular to the direction of the scanning OCT beam. The method uses a concept of structuring, directional kernels of shapes matching the geometry of the image features. We use rotating ellipses to find the most likely local orientation of the vessels and use the best matching ellipses for median filtering of the C-scans. We demonstrate our approach in the imaging of a normal human eye with laboratory-grade spectral-domain OCT setup. The "field performance" is demonstrated in imaging of diabetic retinopathy cases with a commercial OCT device. The absolute complex differences method is used for the generation of OCTA images from the data collected in the most noise-wise unfavorable OCTA scanning regime-two frame scanning.