Segmentation of retinal blood vessels in optical coherence tomography angiography images

1. A method for segmentation of retinal blood vessels comprising:

providing at least one optical coherence tomography angiography digital image, the image including a plurality of pixels;

preprocessing the image, wherein the preprocessing includes calculating a threshold for separating blood vessels from background in the image;

segmenting the image, wherein the segmenting includes calculating a probability, based on at least one probability model, that each pixel in the plurality of pixels depicts a blood vessel or background; and

refining the segmented image;

wherein the refining includes labeling each pixel as depicting either a blood vessel or background based on said probability and, after the labeling, relabeling each pixel as depicting either a blood vessel or background based on equivalence between said pixel and pixels neighboring said pixel.

2. The method of claim 1 , wherein the preprocessing further includes uniformly distributing grey levels across each image.

3. The method of claim 1 , wherein the preprocessing further includes applying a smoothing factor to improve contrast in the images.

4. The method of claim 1 , wherein the preprocessing comprises uniformly distributing grey levels across each image, calculating a threshold for separating blood vessels from background in the images, and applying a smoothing factor to improve contrast in the images.

5. The method of claim 1 , wherein the segmenting includes calculating a probability, based on a plurality of probability models, that each pixel in the plurality of pixels depicts a blood vessel or background.

6. The method of claim 5 , wherein the segmenting includes calculating a probability, based on a joint probability model generated by combining the plurality of probability models, that each pixel in the plurality of pixels depicts a blood vessel or background.

7. The method of claim 6 , wherein the plurality of probability models include a prior gray intensity model, a first order intensity model, and a higher order Markov-Gibbs random field model.

8. The method of claim 1 , wherein the at least one optical coherence tomography angiography digital image is a deep optical coherence tomography angiography digital image and a superficial optical coherence tomography angiography digital image.

9. A method for diagnosing diabetic retinopathy comprising:

preprocessing at least one optical coherence tomography angiography digital image for a patient, the image including a plurality of pixels, wherein the preprocessing includes uniformly distributing grey levels across the image, calculating a threshold for separating blood vessels from background in the image, and applying a smoothing factor to improve contrast in the image;

segmenting the images, wherein the segmenting includes calculating a probability, based on at least one probability model, that each pixel in the plurality of pixels depicts a blood vessel or background;

refining the segmented image; wherein the refining includes labeling each pixel as depicting either a blood vessel or background based on said probability;

extracting retinal features correlated with diabetic retinopathy from the image; and

submitting the extracted retinal features to a supervised learning model trained to classify images as normal state or diabetic retinopathy state;

wherein the supervised learning model outputs a classification of normal state or diabetic retinopathy state for the patient based on the extracted retinal features.

10. The method of claim 9 , wherein the extracted retinal features are at least one of local density of retinal blood vessels, caliber of retinal vessels, and size of the foveal avascular zone.

11. The method of claim 10 , wherein the extracted retinal features are local density of retinal blood vessels, caliber of retinal vessels, and size of the foveal avascular zone.

12. The method of claim 9 , wherein the supervised learning model is a support vector machine classifier with radial basis function kernel.

13. The method of claim 9 , wherein the segmenting includes calculating a probability, based on a joint probability model generated by combining a plurality of probability models, that each pixel in the plurality of pixels depicts a blood vessel or background.

14. An article of manufacture having one or more non-transitory computer readable storage media storing instructions thereon which, when executed by a computing device, cause the device to perform a method comprising:

preprocessing an optical coherence tomography angiography digital image, the image including a plurality of pixels, wherein the preprocessing includes calculating a threshold for separating blood vessels from background in the image;

segmenting the image, wherein the segmenting includes calculating a probability, based on a joint probability model generated by combining a plurality of probability models, that each pixel in the plurality of pixels depicts a blood vessel or background; and

refining the segmented image; wherein the refining includes labeling each pixel as depicting either a blood vessel or background based on said probability.

15. The article of claim 14 , wherein the method further comprises

extracting retinal features correlated with a retinovascular disease from the image; and

submitting the extracted retinal features to a supervised learning model trained to classify images as normal state or disease state;

wherein the supervised learning model outputs a classification of normal state or disease state for the patient based on the extracted retinal features.

16. The article of claim 15 , wherein the retinovascular disease is diabetic retinopathy, and wherein the disease state is diabetic retinopathy state.

17. The article of claim 15 , wherein the extracted retinal features are at least one of local density of retinal blood vessels, caliber of retinal vessels, and size of the foveal avascular zone.