Diagnostic accuracy and clinical performance of deep learning models for grading diabetic retinopathy: a systematic review and meta-analysis

1

Diabetic retinopathy (DR) is a major cause of preventable vision loss, necessitating accurate severity grading for effective management.
2

Deep learning (DL) models have shown high sensitivity in grading DR, particularly excelling in detecting no DR and vision-threatening stages.
3

The systematic review included 41 studies, revealing significant variability in DL model sensitivities across different DR severity levels.
4

In a simplified four-class classification, DL models demonstrated improved sensitivities for all grades of DR compared to the five-class task.
5

The review highlights the need for methodological standardization and external validation to enhance the clinical utility of DL in DR grading.

1