Automated determination of arteriovenous ratio in images of blood vessels

Inventors

Abramoff, Michael D.Niemeijer, MelndertXu, XiayuSonka, MilanReinhardt, Joseph M.

Assignees

Reinhardt Joseph MUnversity Of Iowa Research FoundationUniversity of Iowa Research Foundation UIRFUS Department of Veterans Affairs

Publication Number

US-9924867-B2

Publication Date

2018-03-27

Expiration Date

2032-01-20

Interested in licensing this patent?

MTEC can help explore whether this patent might be available for licensing for your application.


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.

Core Innovation

The invention automatically determines the Arteriolar-to-Venular diameter Ratio (AVR) in images of blood vessels, such as retinal vessels and other blood vessels in vertebrates. The AVR is a critical predictor of risks for stroke, cerebral atrophy, cognitive decline, and myocardial infarct. The methods and systems locate the optic disc, define a region of interest (ROI), classify vessels as arteries or veins, estimate vessel widths, and calculate the AVR. Previously, AVR determination required manual, expert analysis of retinal color fundus images, which was time-consuming and limited to gross estimates rather than numeric ratios.

The problem being solved is the lack of an automated, accurate, and quantitative method for determining AVR from fundus images. Manual approaches are laborious and cannot provide precise numeric AVR values. Clinicians traditionally can identify only grossly abnormal arteriovenous ratios and cannot perform detailed vessel classification or width measurements automatically. This impacts the timely risk analysis for cardiovascular and brain diseases. The invention addresses this by using image preprocessing, vessel segmentation, vessel skeletonization, pixel classification using supervised learning, vessel width measurements via novel approaches including tobogganing and graph-based methods, and iterative vessel pairing to calculate AVR automatically and accurately.

Claims Coverage

The claims cover three independent aspects: a method, a system, and a non-transitory computer readable medium for automatic determination of AVR from images.

Preprocessing and vessel segmentation with trained classifier

Receiving fundus images with a field of view border, applying mirroring of pixel values across the border, removing slow background variations by filtering, then performing vessel segmentation using a trained classifier to distinguish arteries from veins, producing a vessel likelihood map.

Application of tobogganing and skeletonization for vessel delineation

Applying a tobogganing method to create splat maps, determining for each splat whether it lies inside a vessel based on the likelihood map, then skeletonizing the vessel likelihood map and removing bifurcations and crossovers to obtain distinct vessel segments labeled as artery or vein.

Definition and use of region of interest centered on optic disc

Detecting optic disc centerpoint, defining a region of interest (ROI) around it, and restricting analysis and vessel segments to those within this ROI.

Measurement of vessel widths from processed map

For each centerline pixel of vessel segments, determining left and right vessel edges in the processed vessel likelihood map and calculating vessel width as the distance between these edges.

Estimation of the arteriovenous ratio (AVR)

Calculating AVR from vessel width measurements using iterative algorithms, inclusive of estimating central retinal artery equivalent (CRAE) and central retinal vein equivalent (CRVE) diameters at multiple distances and interpolating to final AVR.

Utilization of multiscale and graph-based approaches for vessel width determination

Using graph search with a multiscale cost function derived from wavelet kernel lifting including Gabor, Gaussian derivative, and Difference of Gaussians kernels to obtain accurate vessel boundary segmentation and width measures.

Together, these inventive features enable a fully automated, accurate, and quantitative system and method for AVR determination from retinal or other blood vessel images, improving upon prior manual and less precise methods.

Stated Advantages

Automated AVR determination provides a quantitative approach to early detection and risk analysis for cardiovascular and brain diseases.

The method reduces time consumption and reliance on expert manual measurements.

The system accurately classifies arteries and veins with high accuracy using supervised classification.

The graph-based vessel boundary detection detects both boundaries simultaneously, improving robustness and accuracy even in low-contrast or blurred vessels.

The methods support batch processing for large image sets, enabling population-level disease propensity assessment.

Documented Applications

Risk analysis and prediction of cardiovascular events, stroke, cerebral atrophy, cognitive decline, and myocardial infarct using fundus imaging.

Applications in Medicine, Neurology, Primary Care, Ophthalmology for analysis of retinal blood vessels.

Extension to imaging of blood vessels in other body parts suitable for multi-wavelength imaging: iris, skin, eardrum, brain surface, and other organs in albino animals.

Potential application to two-dimensional blood vessel projections such as cardiac and brain angiograms.

JOIN OUR MAILING LIST

Stay Connected with MTEC

Keep up with active and upcoming solicitations, MTEC news and other valuable information.