Artery vein classification in fundus images using serially connected U-Nets

BACKGROUND AND OBJECTIVE

Retinal vessels provide valuable information when diagnosing or monitoring various diseases affecting the retina and disorders affecting the cardiovascular or central nervous systems. Automated retinal vessel segmentation can assist clinicians and researchers when interpreting retinal images. As there are differences in both the structure and function of retinal arteries and veins, separating these two vessel types is essential. As manual segmentation of retinal images is impractical, an accurate automated method is required.

METHODS

In this paper, we propose a convolutional neural network based on serially connected U-nets that simultaneously segment the retinal vessels and classify them as arteries or veins. Detailed ablation experiments are performed to understand how the major components contribute to the overall system's performance. The proposed method is trained and tested on the public DRIVE and HRF datasets and a proprietary dataset.

RESULTS

The proposed convolutional neural network achieves an F₁ score of 0.829 for vessel segmentation on the DRIVE dataset and an F₁ score of 0.814 on the HRF dataset, consistent with the state-of-the-art methods on the former and outperforming the state-of-the-art on the latter. On the task of classifying the vessels into arteries and veins, the method achieves an F₁ score of 0.952 for the DRIVE dataset exceeding the state-of-the-art performance. On the HRF dataset, the method achieves an F₁ score of 0.966, which is consistent with the state-of-the-art.

CONCLUSIONS

The proposed method demonstrates competitive performance on both vessel segmentation and artery vein classification compared with state-of-the-art methods. The method outperforms human experts on the DRIVE dataset when classifying retinal images into arteries, veins, and background simultaneously. The method segments the vasculature on the proprietary dataset and classifies the retinal vessels accurately, even on challenging pathological images. The ablation experiments which utilize repeated runs for each configuration provide statistical evidence for the appropriateness of the proposed solution. Connecting several simple U-nets significantly improved artery vein classification performance. The proposed way of serially connecting base networks is not limited to the proposed base network or segmenting the retinal vessels and could be applied to other tasks.

Automation improves repeatability of retinal oximetry measurements

Purpose

Retinal oximetry is a technique based on spectrophotometry where images are analyzed with software capable of calculating vessel oxygen saturation and vessel diameter. In this study, the effect of automation of measurements of retinal vessel oxygen saturation and vessel diameter is explored.

Methods

Until now, operators have had to choose each vessel segment to be measured explicitly. A new, automatic version of the software automatically selects the vessels once the operator defines a measurement area. Five operators analyzed image pairs from the right eye of 23 healthy subjects with semiautomated retinal oximetry analysis software, Oxymap Analyzer (v2.5.1), and an automated version (v3.0). Inter- and intra-operator variability was investigated using the intraclass correlation coefficient (ICC) between oxygen saturation measurements of vessel segments in the same area of the retina.

Results

For semiautomated saturation measurements, the inter-rater ICC was 0.80 for arterioles and venules. For automated saturation measurements, the inter-rater ICC was 0.97 for arterioles and 0.96 for venules. For semiautomated diameter measurements, the inter-rater ICC was 0.71 for arterioles and venules. For automated diameter measurements the inter-rater ICC was 0.97 for arterioles and 0.95 for venules. The inter-rater ICCs were different (p < 0.01) between the semiautomated and automated version in all instances.

Conclusion

Automated measurements of retinal oximetry values are more repeatable compared to measurements where vessels are selected manually.

Retinal oxygen metabolism in patients with mild cognitive impairment

Introduction

We have previously reported that retinal vessel oxygen saturation is increased in mild-to-moderate dementia of Alzheimer's type when compared with healthy individuals. Mild cognitive impairment (MCI) is the predementia stage of the disease. The main purpose was to investigate if these changes are seen in MCI.

Methods

Retinal vessel oxygen saturation was measured in 42 patients with MCI and 42 healthy individuals with a noninvasive retinal oximeter, Oxymap T1. The groups were paired according to age.

Results

Arteriolar and venular oxygen saturation was increased in MCI patients compared to healthy individuals (arterioles: 93.1 ± 3.7% vs. 91.1 ± 3.4%, P = .01; venules: 59.6 ± 6.1% vs. 54.9 ± 6.4%, P = .001). Arteriovenous difference was decreased in MCI compared to healthy individuals (33.5 ± 4.5% vs. 36.2 ± 5.2%, P = .01).

Discussion

Increased retinal vessel oxygen saturation and decreased arteriovenous difference in MCI could reflect less oxygen extraction by retinal tissue. This indicates that retinal oxygen metabolism may be affected in patients with MCI.

•

The need for reliable, noninvasive techniques for diagnosis of dementia is widely recognized.

•

This research indicates that retinal metabolism is decreased in patients in the predementia stage of mild cognitive impairment.

•

Retinal oximetry is a novel noninvasive method that could help as a diagnostic tool in dementia.

Automatic registration of retina images based on genetic techniques

The aim of this paper is to develop an automatic method for the registration of multitemporal digital images of the fundus of the human retina. The images are acquired from the same patient at different times by a color fundus camera. The proposed approach is based on the application of global optimization techniques to previously extracted maps of curvilinear structures in the images to be registered (such structures being represented by the vessels in the human retina): in particular, a genetic algorithm is used, in order to estimate the optimum transformation between the input and the base image. The algorithm is tested on two different types of data, gray scale and color images, and for both types, images with small changes and with large changes are used. The comparison between the registered images using the implemented method and a manual one points out that the proposed algorithm provides an accurate registration. The convergence to a solution is not possible only when dealing with images taken from very different view-points.

Automatic Retinal Oximetry

PURPOSE

To measure hemoglobin oxygen saturation (SO(2)) in retinal vessels and to test the reproducibility and sensitivity of an automatic spectrophotometric oximeter.

METHODS

Specialized software automatically identifies the retinal blood vessels on fundus images, which are obtained with four different wavelengths of light. The software calculates optical density ratios (ODRs) for each vessel. The reproducibility was evaluated by analyzing five repeated measurements of the same vessels. A linear relationship between SO(2) and ODR was assumed and a linear model derived. After calibration, reproducibility and sensitivity were calculated in terms of SO(2). Systemic hyperoxia (n = 16) was induced in healthy volunteers by changing the O(2) concentration in inhaled air from 21% to 100%.

RESULTS

The automatic software enhanced reproducibility, and the mean SD for repeated measurements was 3.7% for arterioles and 5.3% venules, in terms of percentage of SO(2) (five repeats, 10 individuals). The model derived for calibration was SO(2) = 125 - 142 . ODR. The arterial SO(2) measured 96% +/- 9% (mean +/- SD) during normoxia and 101% +/- 8% during hyperoxia (n = 16). The difference between normoxia and hyperoxia was significant (P = 0.0027, paired t-test). Corresponding numbers for venules were 55% +/- 14% and 78% +/- 15% (P < 0.0001). SO(2) is displayed as a pseudocolor map drawn on fundus images.

CONCLUSIONS

The retinal oximeter is reliable, easy to use, and sensitive to changes in SO(2) when concentration of O(2) in inhaled air is changed.