An overview of artificial intelligence in diabetic retinopathy and other ocular diseases

Abstract

Artificial intelligence (AI), also known as machine intelligence, is a branch of science that empowers machines using human intelligence. AI refers to the technology of rendering human intelligence through computer programs. From healthcare to the precise prevention, diagnosis, and management of diseases, AI is progressing rapidly in various interdisciplinary fields, including ophthalmology. Ophthalmology is at the forefront of AI in medicine because the diagnosis of ocular diseases heavy reliance on imaging. Recently, deep learning-based AI screening and prediction models have been applied to the most common visual impairment and blindness diseases, including glaucoma, cataract, age-related macular degeneration (ARMD), and diabetic retinopathy (DR). The success of AI in medicine is primarily attributed to the development of deep learning algorithms, which are computational models composed of multiple layers of simulated neurons. These models can learn the representations of data at multiple levels of abstraction. The Inception-v3 algorithm and transfer learning concept have been applied in DR and ARMD to reuse fundus image features learned from natural images (non-medical images) to train an AI system with a fraction of the commonly used training data (<1%). The trained AI system achieved performance comparable to that of human experts in classifying ARMD and diabetic macular edema on optical coherence tomography images. In this study, we highlight the fundamental concepts of AI and its application in these four major ocular diseases and further discuss the current challenges, as well as the prospects in ophthalmology.

Keywords: age-related macular degeneration; artificial intelligence; cataract; diabetic retinopathy; glaucoma.

Figure 1

A diagram illustrating a fast R-CNN algorithm for automatic lesion detection and disease recognition from fundus images. The input fundus image will be fed into the CNN network to get the corresponding feature map. The derived feature map will be used to estimate region proposals (candidate lesion regions in squared boxes), which will then be classified and predicted as different disease categories. FC, fully connected layers; DM, diabetes mellitus; DR, diabetic retinopathy; R-CNN, region-based convolutional neural network; ROI, region of interest.

Figure 2

Using deep residual network (ResNet) for cataract recognition and grading. The overall architecture of the ResNet consists of 16 residual blocks and each residual block consists of three convolutional layers. The output of the ResNet includes: (A1) mode recognition - identify the capture mode between mydriatic and non-mydriatic images, and between optical section and diffuse slit lamp illimitation; (A2) cataract recognition - the system could classify the images as normal (no cataract), cataractous, or postoperative intraocular lens (IOL); and (A3) severity evaluation - classify the type and severity of the cataract (A–G), and assess the subsequent follow-up or referral arrangements for the patient. Conv, convolutional layers.

Figure 3

Conventional framework for ARMD detection from fundus images. (1) Preprocessing – image preprocessing is performed on the input fundus image to reduce noise and enhance image quality. (2) Feature Extraction – image features such as texture, entropy and color features will be extracted from the preprocessed images. (3) Feature Selection – feature selection will be conducted on the extracted image features to select the best representative features of an image. (4) Training – at the training phase, a model such as support vector machine (SVM) will be built that tries to separate the training data into different categories e.g., ARMD and non-ARMD. (5) Testing – testing phase will apply the trained model to unseen fundus images and classify them to the known categories e.g., ARMD and non-ARMD.

Figure 4

Age-related macular degeneration (ARMD) lesions segmentation based on U-Net. U-Net is one of the most widely used segmentation architectures for biomedical images and stemmed from the fully convolutional network. The U-Net model consists of a downsampling path and upsampling path, where downsampling path has convolutional and max-pooling layers to extract high-level abstract information while the upsampling path has convolutional and deconvolutional layers that upsample the feature maps to output the segmentation outcomes. For ARMD segmentation, U-Net will take OCT images as the input and progressively extract semantic features that allow to separate the lesions from the surrounding background and output the lesion segmentation results.