Automated determination of arteriovenous ratio in images of blood vessels

Patent metadata

Field	Value
Publication number	US-9924867-B2
Application number	US-201213355386-A
Country	US
Kind code	B2
Filing date	Jan 20, 2012
Priority date	Jan 20, 2011
Publication date	Mar 27, 2018
Grant date	Mar 27, 2018

Abstract

The methods and systems provided can automatically determine an Arteriolar-to-Venular diameter Ratio, AVR, in blood vessels, such as retinal blood vessels and other blood vessels in vertebrates. The AVR is an important predictor of increases in the risk for stroke, cerebral atrophy, cognitive decline, and myocardial infarct.

First claim

What is claimed is:

1. A method comprising:
receiving, by a computing device, an image, wherein the image comprises a digital color fundus image comprising a border around a field of view;
mirroring one or more pixel values in the image from within the border to outside the field of view;
blurring the image with a filter and subtracting the blurred image from the image to remove slow background variations, resulting in a preprocessed image;
identifying one or more vessel pixels in the preprocessed image as representing arteries and one or more other vessel pixels in the preprocessed image as representing veins in the field of view, by performing vessel segmentation on the preprocessed image and utilizing a trained classifier to disambiguate arterial vessel pixels from venous vessel pixels, wherein the trained classifier is trained based on a set of training images, resulting in a vessel likelihood map comprising a plurality of vessels;
applying a tobogganing method to the preprocessed image, resulting in a splat map;
determining, for each splat of the slap map, based on the vessel likelihood map, whether a splat is inside or outside a vessel, resulting in a processed vessel likelihood map;
applying a skeletonization method to the processed vessel likelihood map to reduce each of the plurality of vessels to a single respective centerline comprised of centerline pixels;
removing cross-over points and bifurcation points of the respective centerlines the processed vessel likelihood map to subdivide the processed vessel likelihood map into a plurality of vessel segments, each labeled as an artery vessel segment or a vein vessel segment;
defining a region of interest in the processed vessel likelihood map centered on a centerpoint of an optic disc;
removing vessels segments from the plurality of vessel segments that are outside the region of interest;
determining, by the computing device, vessel width measurements for the plurality of vessel segments, wherein the vessel width measurement is measured by, for each centerline pixel, determining a left vessel edge and a right vessel edges in the processed vessel likelihood map and calculating the distance between the left vessel edge and the right vessel edge; and
estimating, by the computing device, from the vessel width measurements an arteriovenous ratio (AVR).

2. The method of claim 1 , wherein the trained classifier uses a feature vector comprising one or more of, derivatives, texture, and color properties to disambiguate arterial vessel pixels from venous vessel pixels or arterial vessel segments from venous vessel segments.

3. The method of claim 1 , wherein vessel tree analysis and globally optimal graph search are used to disambiguate arterial vessel pixels from venous vessel pixels or arterial vessel segments from venous vessel segments.

4. The method of claim 1 , wherein blood flow is used to disambiguate arterial vessel pixels from venous vessel pixels or arterial vessel segments from venous vessel segments.

5. The method of claim 1 , wherein a higher propensity for a disease is indicated by a decreased AVR.

6. The method of claim 1 , wherein the image is one or more of, a color image, a multispectral image, an Optical Coherence Tomography image.

7. The method of claim 1 , wherein the image is of the retina, iris, skin, brain surface, or any tissue with visible blood vessels imaged using any two different wavelength imaging processes.

8. The method of claim 1 , wherein determining vessel width measurements for the identified arteries and veins comprises using one or more of, a graph search or profile fitting.

9. The method of claim 8 , wherein determining vessel width measurements for the identified arteries and veins comprises a graph search, wherein the graph search uses a multiscale cost function.

10. The method of claim 9 , wherein the multiscale cost function is derived from wavelet kernel lifting.

11. The method of claim 1 , wherein estimating, from the vessel width measurements, an AVR comprises:
estimating, by the computing device, a plurality of central retinal artery equivalent (CRAE) diameters and a plurality of central retinal vein equivalent (CRVE) diameters at a plurality of distances from an optic disc;
interpolating, by the computing device, a single CRAE and a single CRVE from the estimated plurality of diameters; and
calculating, by the computing device, the AVR from the single CRAE and the single CRVE.

12. A system comprising:
a memory; and
a processor, coupled to the memory, wherein the processor is configured for performing steps comprising,
receiving an image, wherein the image comprises a digital color fundus image comprising a border around a field of view;
mirroring one or more pixel values in the image from within the border to outside the field of view;
blurring the image with a filter and subtracting the blurred image from the image to remove slow background variations, resulting in a preprocessed image;
identifying one or more vessel pixels in the preprocessed image as representing arteries and one or more other vessel pixels in the preprocessed image as representing veins in the field of view, by performing vessel segmentation on the preprocessed image and utilizing a trained classifier to disambiguate arterial vessel pixels from venous vessel pixels, wherein the trained classifier is trained based on a set of training images, resulting in a vessel likelihood map comprising a plurality of vessels;
applying a tobogganing method to the preprocessed image, resulting in a splat map;
determining, for each splat of the slap map, based on the vessel likelihood map, whether a splat is inside or outside a vessel, resulting in a processed vessel likelihood map;
applying a skeletonization method to the processed vessel likelihood map to reduce each of the plurality of vessels to a single respective centerline comprised of centerline pixels;
removing cross-over points and bifurcation points of the respective centerlines the processed vessel likelihood map to subdivide the processed vessel likelihood map into a plurality of vessel segments, each labeled as an artery vessel segment or a vein vessel segment;
defining a region of interest in the processed vessel likelihood map centered on a centerpoint of an optic disc;
removing vessels segments from the plurality of vessel segments that are outside the region of interest;
determining, by the computing device, vessel width measurements for the plurality of vessel segments, wherein the vessel width measurement is measured by, for each centerline pixel, determining a left vessel edge and a right vessel edges in the processed vessel likelihood map and calculating the distance between the left vessel edge and the right vessel edge; and
estimating, by the computing device, from the vessel width measurements an arteriovenous ratio (AVR).

13. The system of claim 12 , wherein higher propensity for a disease is indicated by a decreased AVR.

14. The system of claim 12 , wherein the image is of the retina, iris, skin, brain surface, or any tissue with visible blood vessels imaged using any two different wavelength imaging processes.

15. The system of claim 12 , wherein determining vessel width measurements for the identified arteries and veins comprises a graph search.

16. The system of claim 15 , wherein the graph search uses a multiscale cost function.

17. The system of claim 16 , wherein the multi scale cost function is derived from wavelet kernel lifting comprising Gabor, Gaussian derivative, or difference of Gaussians kernels.

Assignees

• Abramoff Michael D
• Niemeijer Melndert
• Xu Xiayu
• Sonka Milan
• Reinhardt Joseph M
• Univ Iowa Res Found
• The Us Secretary Of The Department Of Veterans Affairs

Inventors

• Abramoff Michael D
• Niemeijer Melndert
• Xu Xiayu
• Sonka Milan
• Reinhardt Joseph M

Classifications

• A61B3/1225

  using coherent radiation · CPC title

• A61B3/0058

  for multiple images · CPC title

• A61B3/12 Primary

  for looking at the eye fundus, e.g. ophthalmoscopes (A61B3/13 takes precedence) · CPC title

• A61B3/0016

  Operational features thereof · CPC title