Fact checked byHeather Biele

Read more

February 21, 2024
1 min read
Save

Automated machine-learning models may help predict diabetic retinopathy progression

Fact checked byHeather Biele
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Key takeaways:

  • The machine-learning model successfully identified all eyes with mild NPDR that progressed within 1 year.
  • The model identified 85% of eyes with moderate NPDR that progressed within 1 year.

Automated machine-learning models accurately identified risk for diabetic retinopathy progression using ultra-widefield retinal images, according to a study published in JAMA Ophthalmology.

“Estimating the risk of diabetic retinopathy (DR) progression is one of the most important and challenging tasks clinicians face when caring for individuals with diabetic eye disease,” Paolo S. Silva, MD, from the Beetham Eye Institute at Joslin Diabetes Center, and colleagues wrote.

eyes
Automated machine-learning models may assist in assessing risk for diabetic retinopathy disease progression, according to recent research. Image: Adobe Stock

In a prospective development and validation study of automated machine-learning models to predict DR progression, researchers included 1,179 deidentified ultra-widefield retinal images, all of which were taken with the California retinal imager (Optos). Of those, 32.2% had mild nonproliferative DR (NPDR), and 67.8% had moderate NDPR.

According to results, half of the training set had DR progression. Using area under the precision-recall curve (AUPRC) to determine model accuracy, researchers reported an AUPRC of 0.717 for mild NPDR with a precision and recall of 53.16% and an AUPRC of 0.863 for moderate NPDR with a precision and recall of 75%.

In the validation set, the model successfully identified 77.5% of eyes with mild NPDR and 85.4% of eyes with moderate NPDR that progressed at least two steps. The model also successfully identified all eyes with mild NPDR and 85% of eyes with moderate NPDR that progressed within 1 year.

“Although prospective validation and regulatory approval is required before these AI models are made available to physicians for clinical use, our results highlight the increasing accessibility of [machine learning] applications to address unmet clinical needs that may improve screening and vision outcomes for patients with diabetes,” Silva and colleagues wrote.