Automatic Segmentation of Retinal Capillaries in Adaptive Optics Scanning Laser Ophthalmoscope Perfusion Images Using a Convolutional Neural Network

Abstract

Adaptive optics scanning laser ophthalmoscope (AOSLO) capillary perfusion images can possess large variations in contrast, intensity, and background signal, thereby limiting the use of global or adaptive thresholding techniques for automatic segmentation. We sought to develop an automated approach to segment perfused capillaries in AOSLO images. Methods: 12,979 image patches were extracted from manually segmented AOSLO montages from 14 eyes and used to train a convolutional neural network (CNN) that classified pixels as capillaries, large vessels, background, or image canvas. 1764 patches were extracted from AOSLO montages of four separate subjects, and were segmented manually by two raters (ground truth) and automatically by the CNN, an Otsu's approach, and a Frangi approach. A modified Dice coefficient was created to account for slight spatial differences between the same manually and CNN-segmented capillaries. Results: CNN capillary segmentation had an accuracy (0.94), a Dice coefficient (0.67), and a modified Dice coefficient (0.90) that were significantly higher than other automated approaches (P < 0.05). There were no significant differences in capillary density and mean segment length between manual ground-truth and CNN segmentations (P > 0.05). Conclusions: Close agreement between the CNN and manual segmentations enables robust and objective quantification of perfused capillary metrics. The developed CNN is time and computationally efficient, and distinguishes capillaries from areas containing diffuse background signal and larger underlying vessels. Translational Relevance: This automatic segmentation algorithm greatly increases the efficiency of quantifying AOSLO capillary perfusion images.

Longitudinal in vivo changes in radial peripapillary capillaries and optic nerve head structure in non-human primates with early experimental glaucoma

“…35,[48][49][50] The average mean MRW for all control eyes across all time-points in this study (340.0 ± 50.7 µm) is also similar to the mean value previously reported in control eyes by Ivers et al 35 (308.7 ± 55.1 µm) and is within the range reported by Strouthidis et al (~200 -425 µm). 44 In addition, the mean global RNFLT measured in the control eyes over the course of this study (116.4 ± 6.8 µm) is within the range of values of global RNFLT published in healthy NHP eyes (~101 -124 µm). 35,39,51,52 Coefficients of variation (CoVs) computed using the control eye data for all OCT-derived parameters in this study were also consistent with CoVs for the same metrics reported by other studies.…”

“…RPC montages were then segmented with a convolutional neural network (CNN) that has been previously described. 44…”

AOSLO-net: A Deep Learning-Based Method for Automatic Segmentation of Retinal Microaneurysms From Adaptive Optics Scanning Laser Ophthalmoscopy Images

“…2 ), to perform automatic segmentation of MAs from AOSLO images and quantify their shape metrics that can be used for classification of MAs into different types, such as focal bulging, saccular, fusiform, mixed saccular/fusiform, pedunculated, and irregular-shaped MAs. 14 The objective of this work is fundamentally different from some recent studies, which either use the AOSLO technique to segment microvessels without MAs 34 , 35 or focus on segmenting MAs from fundus images, 22 , 36 , 37 which do not provide the resolution required to identify the shape of MAs. This model is trained and tested by using 87 AOSLO MA images with masks generated manually by ophthalmologists or trained graders, the largest published AOSLO image dataset for this kind of effort thus far.…”

Artificial Intelligence Segmented Dynamic Video Images for Continuity Analysis in the Detection of Severe Cardiovascular Disease

“…The covert nature of the disease determines that ECG cannot be observed only, and more accurate medical images are needed to discover hidden diseases and symptoms, to carry out timely intervention and treatment; the sudden nature of the disease also determines that medical images cannot be used only, and ECG needs to be portable to monitor the condition of the heart in real-time and provide timely warning of possible dangers ( Randive et al, 2020 ). Only when ECG and medical imaging cooperate, can the mortality rate of cardiovascular disease be effectively reduced ( Musial et al, 2020 ). However, the existing dynamic ECG equipment still adopts offline storage mode, which cannot realize timely signal analysis and early warning; medical image analysis is limited by the professional level of doctors themselves, and at the same time is easily interfered by other factors, which can easily produce inconsistent results among different doctors, and cannot achieve the purpose of accurate diagnosis and treatment ( Aminikhanghahi and Cook, 2017 ).…”