Absolute retinal blood flow in healthy eyes and in eyes with retinal vein occlusion.

PURPOSE: To measure non-invasively retinal venous blood flow (RBF) in healthy subjects and patients with retinal venous occlusion (RVO). METHODS: The prototype named AO-LDV (Adaptive Optics Laser Doppler Velocimeter), which combines a new absolute laser Doppler velocimeter with an adaptive optics fundus camera (rtx1, Imagine Eyes®, Orsay, France), was studied for the measurement of absolute RBF as a function of retinal vessel diameters and simultaneous measurement of red blood cell velocity. RBF was measured in healthy subjects (n = 15) and patients with retinal venous occlusion (RVO, n = 6). We also evaluated two softwares for the measurement of retinal vessel diameters: software 1 (automatic vessel detection, profile analysis) and software 2 (based on the use of deep neural networks for semantic segmentation of vessels, using a M2u-Net architecture). RESULTS: Software 2 provided a higher rate of automatic retinal vessel measurement (99.5 % of 12,320 AO images) than software 1 (64.9 %) and wider measurements (75.5 ± 15.7 µm vs 70.9 ± 19.8 µm, p < 0.001). For healthy subjects (n = 15), all the retinal veins in one eye were measured to obtain the total RBF. In healthy subjects, the total RBF was 37.8 ± 6.8 µl/min. There was a significant linear correlation between retinal vessel diameter and maximal velocity (slope = 0.1016; p < 0.001; r2 = 0.8597) and a significant power curve correlation between retinal vessel diameter and blood flow (3.63 × 10-5 × D2.54; p < 0.001; r2 = 0.7287). No significant relationship was found between total RBF and systolic and diastolic blood pressure, ocular perfusion pressure, heart rate, or hematocrit. For RVO patients (n = 6), a significant decrease in RBF was noted in occluded veins (3.51 ± 2.25 µl/min) compared with the contralateral healthy eye (11.07 ± 4.53 µl/min). For occluded vessels, the slope between diameter and velocity was 0.0195 (p < 0.001; r2 = 0.6068) and the relation between diameter and flow was Q = 9.91 × 10-6 × D2.41 (p < 0.01; r2 = 0.2526). CONCLUSION: This AO-LDV prototype offers new opportunity to study RBF in humans and to evaluate treatment in retinal vein diseases.

State-of-the-art retinal vessel segmentation with minimalistic models.

The segmentation of retinal vasculature from eye fundus images is a fundamental task in retinal image analysis. Over recent years, increasingly complex approaches based on sophisticated Convolutional Neural Network architectures have been pushing performance on well-established benchmark datasets. In this paper, we take a step back and analyze the real need of such complexity. We first compile and review the performance of 20 different techniques on some popular databases, and we demonstrate that a minimalistic version of a standard U-Net with several orders of magnitude less parameters, carefully trained and rigorously evaluated, closely approximates the performance of current best techniques. We then show that a cascaded extension (W-Net) reaches outstanding performance on several popular datasets, still using orders of magnitude less learnable weights than any previously published work. Furthermore, we provide the most comprehensive cross-dataset performance analysis to date, involving up to 10 different databases. Our analysis demonstrates that the retinal vessel segmentation is far from solved when considering test images that differ substantially from the training data, and that this task represents an ideal scenario for the exploration of domain adaptation techniques. In this context, we experiment with a simple self-labeling strategy that enables moderate enhancement of cross-dataset performance, indicating that there is still much room for improvement in this area. Finally, we test our approach on Artery/Vein and vessel segmentation from OCTA imaging problems, where we again achieve results well-aligned with the state-of-the-art, at a fraction of the model complexity available in recent literature. Code to reproduce the results in this paper is released.

Competitive neural layer-based method to identify people with high risk for diabetic foot.

BACKGROUND AND OBJECTIVE: To automatically identify patients with diabetes mellitus (DM) who have high risk of developing diabetic foot, via an unsupervised machine learning technique. METHODS: We collected a new database containing 54 known risk factors from 250 patients diagnosed with diabetes mellitus. The database also contained a separate validation cohort composed of 73 subjects, where the perceived risk was annotated by expert nurses. A competitive neuron layer-based method was used to automatically split training data into two risk groups. RESULTS: We found that one of the groups was composed of patients with higher risk of developing diabetic foot. The dominant variables that described group membership via our method agreed with the findings from other studies, and indicated a greater risk for developing such a condition. Our method was validated on the available test data, reaching 71% sensitivity, 100% specificity, and 90% accuracy. CONCLUSIONS: Unsupervised learning may be deployed to screen patients with diabetes mellitus, pointing out high-risk individuals who require priority follow-up in the prevention of diabetic foot with very high accuracy. The proposed method is automatic and does not require clinical examinations to perform risk assessment, being solely based on the information of a questionnaire answered by patients. Our study found that discriminant variables for predicting risk group membership are highly correlated with expert opinion.