Artificial intelligence for early detection of diabetes mellitus complications via retinal imaging

Abstract

Background: Diabetes mellitus (DM) increases the risk of vascular complications, and retinal vasculature imaging serves as a valuable indicator of both microvascular and macrovascular health. Moreover, artificial intelligence (AI)-enabled systems developed for high-throughput detection of diabetic retinopathy (DR) using digitized retinal images have become clinically adopted. This study reviews AI applications using retinal images for DM-related complications, highlighting advancements beyond DR screening, diagnosis, and prognosis, and addresses implementation challenges, such as ethics, data privacy, equitable access, and explainability.

Methods: We conducted a thorough literature search across several databases, including PubMed, Scopus, and Web of Science, focusing on studies involving diabetes, the retina, and artificial intelligence. We reviewed the original research based on their methodology, AI algorithms, data processing techniques, and validation procedures to ensure a detailed analysis of AI applications in diabetic retinal imaging.

Results: Retinal images can be used to diagnose DM complications including DR, neuropathy, nephropathy, and atherosclerotic cardiovascular disease, as well as to predict the risk of cardiovascular events. Beyond DR screening, AI integration also offers significant potential to address the challenges in the comprehensive care of patients with DM.

Conclusion: With the ability to evaluate the patient's health status in relation to DM complications as well as risk prognostication of future cardiovascular complications, AI-assisted retinal image analysis has the potential to become a central tool for modern personalized medicine in patients with DM.

Keywords: Artificial intelligence; Diabetes complications; Diabetes mellitus; Diabetic retinopathy; Retina.

Fig. 1

Illustration of various diabetic complications diagnosed by retinal imaging using AI. (Graphical abstract)

Fig. 2

Schematic representation of fundoscopy showing the DR grading

Fig. 3

The diagram provides a structured overview of the integration of AI in diabetic care, showing the pathways in which AI has significant applications, including DR, nephropathy, neuropathy, cerebrovascular disease, peripheral arterial disease, and cardiovascular assessment. Each branch highlights the potential of AI to enhance diagnosis, monitoring, and management within specialized areas of diabetic care

Fig. 4

A detailed look at the architecture of a convolutional neural network (CNN) for image analysis. a: The convolutional layers (green) use kernels that slide over the three-channel RGB image to recognize key features from the input image. Following the convolution process, the Rectified Linear Unit (ReLU) activation function (green) is applied to introduce non-linearity, enhancing the network’s capability to learn intricate patterns. Subsequently, the max pooling process (orange) is applied, reducing the spatial dimensions by selecting the maximum value within specified regions. b: this part is the multiple repetitions of convolution, ReLU activation function, and max pooling processes, creating the final feature map. c: shows flattening of the last max pooling layer, which converts the 2D feature maps into a 1D vector to prepare the data for the upcoming fully connected layers. d: the architecture transitions to fully connected layers, leading to a classification process where the features are used to provide definitive conclusions

Fig. 5

Performance comparison of AI models across the best metrics for DR screening and diagnosis, as obtained from the included studies. All AUC values are multiplied by 100

Fig. 6

Comparison of AI model performance across various metrics for diabetic complication diagnosis, based on data extracted from the included studies

Fig. 7

This schematic illustrates the interconnected components of a cloud-based system for DR screening. The cloud infrastructure, depicted as remote servers and databases, processes retinal image data captured by a retinal imaging device, such as a fundus camera or OCT scanner. AI algorithms analyze the images within the cloud, generating diagnostic results like risk scores for DR. These results are accessible to healthcare providers through interfaces on computers or mobile devices, ensuring prompt patient care. Patient data privacy measures safeguard sensitive information, including encryption and secure transmission protocols. Additionally, a feedback loop may exist, where diagnostic results contribute to the continuous improvement of AI algorithms over time