Artificial Intelligence and Diabetic Retinopathy: AI Framework, Prospective Studies, Head-to-head Validation, and Cost-effectiveness

Abstract

Current guidelines recommend that individuals with diabetes receive yearly eye exams for detection of referable diabetic retinopathy (DR), one of the leading causes of new-onset blindness. For addressing the immense screening burden, artificial intelligence (AI) algorithms have been developed to autonomously screen for DR from fundus photography without human input. Over the last 10 years, many AI algorithms have achieved good sensitivity and specificity (>85%) for detection of referable DR compared with human graders; however, many questions still remain. In this narrative review on AI in DR screening, we discuss key concepts in AI algorithm development as a background for understanding the algorithms. We present the AI algorithms that have been prospectively validated against human graders and demonstrate the variability of reference standards and cohort demographics. We review the limited head-to-head validation studies where investigators attempt to directly compare the available algorithms. Next, we discuss the literature regarding cost-effectiveness, equity and bias, and medicolegal considerations, all of which play a role in the implementation of these AI algorithms in clinical practice. Lastly, we highlight ongoing efforts to bridge gaps in AI model data sets to pursue equitable development and delivery.

Graphical abstract

Figure 1

Schematic showing the hierarchical organization of AI, machine learning, and deep learning. The panel shows an example of how the deep learning model would be trained on an example data set of 500 images. Each image in the training data set is input into the deep learning model to generate a prediction, and that prediction is compared against the label to calculate the loss. The loss is then used to update the weights of the deep learning model. In this example, this process is repeated for 1,000 epochs, or cycles, in which the model is fed the entire training data set of 500 images.

Figure 2

Example workflow of using AI model for DR screening. A patient would receive a retinal fundus photograph with a clinic-based camera, and the results would be uploaded to the cloud where an AI model would determine gradability of the image and give DR classification and the presence/absence of DME. Back in the clinic, a provider would receive the diagnosis from the AI model and counsel the patient regarding results and discuss referral if indicated.

Figure 3

Left panel: multiple models (1–4) are evaluated on different data sets (A, B, C, and D); thus, the results (1–4) are not comparable. Right panel: models (1–4) are evaluated on data set E, which allows for direct model comparison with an example receiver operating characteristic curve plot. AUC, area under the receiver operating characteristic curve.

Figure 4

All of the different data modalities that will be prospectively collected in the AI-READI study and the possible AI models that can be trained from these data. ECG, electrocardiogram.