Systems and methods of choroidal neovascularization detection using optical coherence tomography angiography

Abstract

Disclosed are systems and methods to automatically detect choroidal neovasculatization (CNV) in the outer retina using OCT angiography. Further disclosed are methods of removing projection artifacts from the outer retina and for combining brightness, orientation, and position information in a context-aware saliency model to quantify CNV area in OCT angiograms.

1 . A method of measuring choroidal neovascularization (CNV) area in an OCT angiogram comprising:
 receiving a set of OCT angiograms;   separating the set of cross sectional-OCT angiograms into an inner set of retina angiograms, the inner set of retina angiograms comprising angiograms located between an internal limiting membrane and an outer plexiform layer, and an outer set of retina angiograms, the outer set of retina angiograms comprising angiograms located between the outer plexiform layer and Bruch's membrane;   projecting the maximum flow values of the inner set of retina angiograms along the axial (Z) direction onto an X-Y plane, thereby generating a single 2D en face inner retina angiogram;   projecting the maximum flow values along the axial (Z) direction of the outer set of retina angiograms onto an X-Y plane, thereby generating a single 2D en face outer retina angiogram;   applying a denoising filter to the inner retina angiogram, thereby generating a denoised inner retina angiogram;   subtracting the denoised inner retina angiogram from the outer retina angiogram;   thereby removing projection artifacts from the outer retina angiogram;   generating a saliency map of CNV in the outer retina angiogram using pattern recognition;   generating a binary vascular mask from the saliency map;   multiplying the binary vascular mask by the original 2D en face outer retina angiogram, thereby identifying the CNV region in the outer retina angiogram; and   calculating the CNV area.   
     
     
         2 . The method of  claim 1  wherein the denoising filter comprises a 20×20 pixel Gaussian filter or low pass filter. 
     
     
         3 . The method of  claim 1  wherein the generation of a saliency map by pattern recognition comprises a calculation of a saliency model that incorporates brightness, orientation contrast, and positional distance as calculated by Equations (1) and (2). 
     
     
         4 . The method of  claim 3  further comprising multiscale enhancement of a saliency model by calculating the saliency map at multiple scales using Equations (3) and (4) and combining the saliency maps according to Equation (5). 
     
     
         5 . The method of  claim 3  further comprising modification of the saliency map to simulate context-awareness according to Equation 6. 
     
     
         6 . The method of  claim 1  wherein generating a vascular mask from the saliency map comprises application of a Laplacian edge detection filter. 
     
     
         7 . The method of  claim 6  further comprising the use of a nonlinear filter to smooth within the target region and preserve the boundary, thereby producing an edge-enhanced saliency map. 
     
     
         8 . The method of  claim 6  further comprising thresholding the edge-enhanced saliency map to extract a rough CNV region. 
     
     
         9 . The method of  claim 8  wherein the thresholding comprises use of Otsu's method. 
     
     
         10 . The method of  claim 6  further comprising application of morphological operators to a rough CNV region to remove small areas and fill holes.