Artificial Intelligence in Diabetic Retinopathy: A Comprehensive Overview

0
0
0

Artificial Intelligence in Diabetic Retinopathy: A Comprehensive Overview

Diabetes is a global health concern, affecting millions. Diabetic retinopathy (DR), a major complication, can lead to blindness. Early detection is crucial, and artificial intelligence (AI) offers new hope.

Introduction

  • Diabetes Prevalence: In 2019, around 463 million adults aged 20 – 79 had diabetes. By 2040, it’s projected to reach 600 million. In China, the prevalence has risen steadily.
  • DR Impact: DR is a leading cause of blindness. It’s progressive, with early stages often asymptomatic. Many patients don’t screen until vision is impaired.

Current Status of DR Screening

  • Guidelines: The American Academy of Ophthalmology recommends annual screenings for type 1 diabetes (5 years after onset) and type 2 (at diagnosis).
  • Screening Programs: In the UK, extensive screening has reduced DR – related blindness. However, compliance is poor globally. Reasons include lack of understanding, resource accessibility, and insurance coverage.
  • Telemedicine: It improves accessibility but still relies on human grading. AI can enhance efficiency.

AI in DR Screening

  • AI Basics: AI, including machine learning (ML) and deep learning (DL), has advanced. Convolutional neural networks (CNNs) are used for image processing. Transfer learning helps models generalize.
  • AI Applications: IDx – DR, EyeArt, and others are approved. AI systems are efficient, accurate, and reduce workload. For example, RetMarker cuts manual grading by 48.42%.

Development of AI – based Diagnostic Systems for DR

  • Dataset Requirements: Training, validation, and test sets should be distinct. Training sets need high – quality, labeled images. Current datasets have limitations like single – source or insufficient non – readable images.
  • Algorithm Development: IDP was early, but IDx – DR (with CNN) improves specificity. EyeArt works on smartphones. AI systems combine with OCT for better DME detection.

Efficacy of Existing AI – based Diagnostic Systems for DR

  • Performance: IDx – DR shows good sensitivity and specificity. EyeArt is feasible for smartphone – based detection. Combining AI and manual work (adjusting thresholds) improves efficiency.
  • Limitations:
    • Dataset Issues: Online datasets may not fit real – world cases. Lack of unified standards and multi – center research questions precision.
    • Classification Standards: ICDR is common, but other standards like early treatment DR study may be better. Different standards affect algorithm validity.
    • Evaluation: No unified testing set. “Black box” phenomenon (inexplicability) and liability issues exist. Information security and single – disease detection are also concerns.

Conclusions and Prospects

  • Future Directions: More AI systems on portable devices (e.g., smartphones) will improve accessibility. Integration with new imaging techniques (multispectral, OCT) will enhance accuracy. AI can support diagnosis, potentially exceeding human accuracy.
  • General Outlook: AI has great prospects. Using clinical resources, creating heterogeneous datasets, and improving standards will widen application. Combining AI with manual work is practical initially.

In conclusion, AI is a powerful tool in DR screening and diagnosis. Despite challenges, its potential to improve patient outcomes and healthcare efficiency is significant.

doi.org/10.1097/CM9.0000000000001816

Was this helpful?

0 / 0