Family Medicine Doctors, AI Could Be Effective in Screening DR

Posters presented at AAO 2022 found that family medicine and artificial intelligence (AI) clinicians were able to screen patients for diabetic retinopathy (DR) effectively, despite the need to improve on false negatives in AI.

In 2 posters presented at AAO 2022, research found that alternatives to ophthalmologists’ screening for diabetic retinopathy (DR) were effective. Clinicians of family medicine and artificial intelligence (AI) accurately identified DR in images of patients’ eyes for diagnostic reasons and for inclusion in clinical trials, respectively.

first poster1 Focus on family medicine physicians’ ability to diagnose DR in a telemedicine program. Telemedicine programs were previously used to effectively screen DR to identify patients in need of eye care. Family medicine physicians evaluate images in telemedicine programs, and ophthalmologists are required only if a diagnosis has not been reached. The researchers aimed to evaluate the effectiveness of family medicine physicians in diagnosing DR in the patients they had seen.

There were 2,260 patients with type 2 diabetes included in this study. They were all photographed within one year with a non-pupillary fundus camera. There were 5 masked family medicine doctors and a retinal specialist who looked at the images and the agreement was compared between the family medicine doctors and the retina specialist.

Patients without apparent retinopathy and no concomitant risk factors will have a repeat retinal imaging within two years. If they do not have apparent retinopathy with risk factors or mild retinopathy without risk factors, then the retinal imaging will be repeated within a year. Patients with mild retinopathy with risk factors will have repeated retinal imaging within 6 months. The Department of Ophthalmology was referred to the Department of Ophthalmology, moderate non-proliferative (NP) DR, proliferative DR, suspected diabetic macular edema, or other optic nerve or retinal hypoplasia.

There were 14 patients with mild NPDR and 27 patients (1.19%) with moderate or severe NPDR in the primary care setting, all of which were confirmed in the ophthalmic setting. There were 180 patients referred for further evaluation, 42 of whom were correctly classified as ‘without DR’, 51 as ‘questionable cases’, 83 as ‘unreadable’ and 4 were classified as false positives as determined Ophthalmologist that there is no DR. .

The authors concluded that family medicine physicians were able to detect 100% of patients with moderate or severe NPDR and were able to correctly classify 93.6% of the retinograms obtained in the telemedicine program, resulting in an excellent performance from the physicians. family Medicine.

second poster2 Focus on using artificial intelligence to identify patients with DR for clinical trials. The U.S. Food and Drug Administration approves the use of artificial intelligence to screen patients with a single macula-centric image in order to identify those who could be transmittable retinopathy. Examination of clinical trials involving DR patients necessitates 47/53 patients on the Early Treatment Diabetic Retinopathy Study Scale using the seven-field holographic protocol to narrow down the pool of potential participants.

There was an estimated 50% screen failure rate in images submitted to the Wisconsin Reading Center. The study aims to determine if artificial intelligence can help screen patients for inclusion in clinical trials.

Artificial intelligence used Field 2 to examine patients. Patients deemed ineligible by AI were not subject to further review. Those deemed eligible by AI went to a human corrector, who would review the seven imaging areas to confirm eligibility or deny eligibility.

The AI ​​was found with an accuracy of 86.4% with a sensitivity of 0.77 and a specificity of 0.89. The AI ​​had a false positive rate of 10.8% and a false negative rate of 22.6%, which was a concern in this screening model. The F1 score for artificial intelligence was 0.72 and the accuracy was found to be 66.7%.

The researchers found that false negatives occurred in AI due to an imbalance of pathology between the central and peripheral fields with more pathology in the peripheral fields. False positives occurred due to pathology imbalance between the central and peripheral domains with more pathology in the central field.

The researchers concluded that the AI ​​algorithm can identify patients with DR levels below 47 but more work needs to be done to reduce the number of false negatives. AI prescreening of eligible patients for DR clinical trials before classifier confirmation can reduce screen failure rate, create cost-effectiveness, and reduce the burden on participants and clinical site staff.

Automated assessment of eligible patients can also improve overall enrollment in Dr.’s clinical trials. More research will be needed to determine this with an external source. The researchers suggest using a prospective clinical trial that would compare the traditional class-one method using a clinical trial in which patients were pre-screened with artificial intelligence.

These two posters show that DR can be screened in multiple ways in the future, where family medicine doctors and AI can help screen patients with DR, which can help treat patients and enroll in clinical trials on DR therapies.


1. Ferreras A, Pinilla I, Abecia E, Figus M, Fogagnolo P, Iester M. The ability of family medicine clinicians to detect DR in a telemedicine program. Filed On: AAO 2022; September 30 – October 3 2022; Chicago, IL. Abstract PO110.

2. Domalpally A, Slater R, Barrett N, Channa R, Blodi B. Artificial intelligence-assisted prescreening of clinical trials of DR. Filed On: AAO 2022; September 30 – October 3 2022; Chicago, IL. Abstract PO341.

Leave a Comment