Improving diabetic retinopathy screening using artificial intelligence: design, evaluation and before-and-after study...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) screening programs are crucial for preventing severe vision loss.
• General practitioners (GPs) at the University Hospital of Navarre (HUN) currently grade fundus images for DR screening.
• Patients requiring further examination are referred to an ophthalmologist.

Purpose of the Study:

• To introduce NaIA-RD, a custom AI tool developed to assist GPs in DR screening.
• To detail the implementation of NaIA-RD within the HUN screening program.
• To evaluate the impact of NaIA-RD on DR screening effectiveness and workflow.

Main Methods:

• A before-and-after study was conducted, comparing 19,828 patients screened pre-NaIA-RD with 22,962 patients screened post-implementation.
• NaIA-RD integrates DR and retinal image quality grading into a single system.
• The study analyzed changes in GP screening criteria and agreement between NaIA-RD and GPs.

Main Results:

• NaIA-RD influenced the screening criteria of 75% of GPs, increasing sensitivity.
• High agreement was observed between NaIA-RD and GPs for non-referral decisions (≥94.6%).
• In autonomous mode, NaIA-RD could reduce workload by 4.27 times without missing sight-threatening DR cases.

Conclusions:

• DR screening is more effective with AI support, such as NaIA-RD.
• NaIA-RD can be safely used for autonomous first-level DR screening.
• Seamless integration of AI tools into clinical workflows can enhance long-term clinical pathways.