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

Artificial intelligence in glaucoma: opportunities, challenges, and future directions

Artificial intelligence (AI) has shown excellent diagnostic performance in detecting various complex problems related to many areas of healthcare including ophthalmology. AI diagnostic systems developed from fundus images have become state-of-the-art tools in diagnosing retinal conditions and glaucoma as well as other ocular diseases. However, designing and implementing AI models using large imaging data is challenging. In this study, we review different machine learning (ML) and deep learning (DL) techniques applied to multiple modalities of retinal data, such as fundus images and visual fields for glaucoma detection, progression assessment, staging and so on. We summarize findings and provide several taxonomies to help the reader understand the evolution of conventional and emerging AI models in glaucoma. We discuss opportunities and challenges facing AI application in glaucoma and highlight some key themes from the existing literature that may help to explore future studies. Our goal in this systematic review is to help readers and researchers to understand critical aspects of AI related to glaucoma as well as determine the necessary steps and requirements for the successful development of AI models in glaucoma.

Intraretinal pigmented cells in retinal degenerative disease

PURPOSE

Invasion of pigmented cells into the retina occurs in retinal degenerative diseases, such as macular telangiectasia type 2 (MacTel) and retinitis pigmentosa (RP). These intraretinal pigmented cells may be derived from the retinal pigment epithelium (RPE), but differences and similarities between intraretinal pigmented cells and RPE have so far not been well characterised.Clinicopathologic case report.

METHOD

Here, we compared intraretinal pigment cells with RPE cells by immunohistochemistry. Immunohistological stains for classic RPE markers (RPE65, CRALBP and KRT18) and blood vessel markers (lectin and collagen 4) were done on sections from postmortem eye tissue from two MacTel donors, an RP donor and a control donor.

MAIN OUTCOME MEASURES

Presence of specific immunohistochemistry markers on intraretinal pigmented and RPE cells.

RESULTS

We found that intraretinal pigmented cells did not express RPE65 and CRALBP, with a small subset expressing them weakly. However, they all expressed KRT18, which was also present in normal RPE cells. Interestingly, we also found clusters of KRT18-positive cells in the retina that were not pigmented.

CONCLUSIONS

Our findings suggest that RPE cells invading the retina dedifferentiate (losing classic RPE markers) and can be pigmented or unpigmented. Therefore, the number of RPE cells invading the retina in retinal degenerative disease may be underappreciated by funduscopy.

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

Purpose

Accurate segmentation of microaneurysms (MAs) from adaptive optics scanning laser ophthalmoscopy (AOSLO) images is crucial for identifying MA morphologies and assessing the hemodynamics inside the MAs. Herein, we introduce AOSLO-net to perform automatic MA segmentation from AOSLO images of diabetic retinas.

Method

AOSLO-net is composed of a deep neural network based on UNet with a pretrained EfficientNet as the encoder. We have designed customized preprocessing and postprocessing policies for AOSLO images, including generation of multichannel images, de-noising, contrast enhancement, ensemble and union of model predictions, to optimize the MA segmentation. AOSLO-net is trained and tested using 87 MAs imaged from 28 eyes of 20 subjects with varying severity of diabetic retinopathy (DR), which is the largest available AOSLO dataset for MA detection. To avoid the overfitting in the model training process, we augment the training data by flipping, rotating, scaling the original image to increase the diversity of data available for model training.

Results

The validity of the model is demonstrated by the good agreement between the predictions of AOSLO-net and the MA masks generated by ophthalmologists and skillful trainees on 87 patient-specific MA images. Our results show that AOSLO-net outperforms the state-of-the-art segmentation model (nnUNet) both in accuracy (e.g., intersection over union and Dice scores), as well as computational cost.

Conclusions

We demonstrate that AOSLO-net provides high-quality of MA segmentation from AOSLO images that enables correct MA morphological classification.

Translational Relevance

As the first attempt to automatically segment retinal MAs from AOSLO images, AOSLO-net could facilitate the pathological study of DR and help ophthalmologists make disease prognoses.

Longitudinal In Vivo Changes in Radial Peripapillary Capillaries and Optic Nerve Head Structure in Non-Human Primates With Early Experimental Glaucoma

Purpose

There is conflicting evidence regarding whether a loss of radial peripapillary capillaries (RPCs) precedes neuronal loss in glaucoma. We examined the time course of in vivo changes in RPCs, optic nerve head (ONH) structure, and retinal nerve fiber layer thickness (RNFLT) in experimental glaucoma (EG).

Methods

Spectral domain optical coherence tomography images were acquired before and approximately every two weeks after inducing unilateral EG in nine rhesus monkeys to quantify mean anterior lamina cribrosa surface depth (ALCSD), minimum rim width (MRW), and RNFLT. Perfused RPC density was measured from adaptive optics scanning laser ophthalmoscope images acquired on the temporal half of the ONH. The time of first significant change was quantified as when values fell and remained outside of the 95% confidence interval established from control eyes.

Results

Mean ALCSD and/or MRW were the first parameters to change in eight EG eyes. RPC density changed first in the ninth. At their first points of change, mean ALCSD posteriorly deformed by 100.2 ± 101.2 µm, MRW thinned by 82.3 ± 65.9 µm, RNFLT decreased by 25 ± 14 µm, and RPC density decreased by 4.5 ± 2.1%. RPC density decreased before RNFL thinning in 5 EG eyes. RNFLT decreased before RPC density decreased in two EG eyes, whereas two EG eyes had simultaneous changes.

Conclusions

In most EG eyes, RPC density decreased before (or simultaneous with) a change in RNFLT, suggesting that vascular factors may play a role in axonal loss in some eyes in early glaucoma.

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

In this paper, an artificial intelligence segmented dynamic video image based on the process of intensive cardiovascular and cerebrovascular disease monitoring is deeply investigated, and a sparse automatic coding deep neural network with a four layers stack structure is designed to automatically extract the deep features of the segmented dynamic video image shot, and six categories of normal, atrial premature, ventricular premature, right bundle branch block, left bundle branch block, and pacing are achieved through hierarchical training and optimization. Accurate recognition of heartbeats with an average accuracy of 99.5%. It provides technical assistance for the intelligent prediction of high-risk cardiovascular diseases like ventricular fibrillation. An intelligent prediction algorithm for sudden cardiac death based on the echolocation network was proposed. By designing an echolocation network with a multilayer serial structure, an intelligent distinction between sudden cardiac death signal and non-sudden death signal was realized, and the signal was predicted 5 min before sudden death occurred, with an average prediction accuracy of 94.32%. Using the self-learning capability of stack sparse auto-coding network, a large amount of label-free data is designed to train the stack sparse auto-coding deep neural network to automatically extract deep representations of plaque features. A small amount of labeled data then introduced to micro-train the entire network. Through the automatic analysis of the fiber cap thickness in the plaques, the automatic identification of thin fiber cap-like vulnerable plaques was achieved, and the average overlap of vulnerable regions reached 87%. The overall time for the automatic plaque and vulnerable plaque recognition algorithm was 0.54 s. It provides theoretical support for accurate diagnosis and endogenous analysis of high-risk cardiovascular diseases.