Read more

November 08, 2023
1 min read
Save

Deep learning algorithm shows utility in identifying DME

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.

The rise of deep learning and artificial intelligence in retinal disease, and all of ophthalmology, is proceeding at a staggering pace.

The innovations over the last several years have laid the foundation for a revolution in the way we care for patients. Use cases for AI and deep learning (DL) include image analysis, disease progression or treatment response prediction, interaction with patients and dissemination of information to patients, and treatment optimization. In clinical trials, AI and DL are poised to improve the entire process of understanding how new medicines affect patients and alter pathology.

Example case with diabetic macular edema
Figure 1. Example case with diabetic macular edema.

Source: George Magrath, MD

Our work to date includes the creation of a DL algorithm to predict the presence of diabetic macular edema based on a single OCT raster scan. Approximately 6,000 patients were included in the project. The five raster scans through the fovea (two above and two below the center foveal scan) were labeled as either “DME” or “no DME.” No other qualifiers were used, and concomitant pathology was not excluded or used as a qualifier. The DL algorithm was then asked to create a predictive model of the presence of DME in these images. Once the algorithm was created, it was tested on images of several hundred additional patients. The algorithm was given only a single OCT raster, without any qualifying information. In this test, the algorithm correctly predicted the presence of DME in more than 95% of patients. Importantly, it also predicted the absence of DME in 95% of patients without DME.

George Magrath, MD
George Magrath

In the clinical trial setting, DL algorithms such as the above are useful in the identification of patients who may benefit from inclusion in a trial from large data sets, or they may lead to more efficient evaluation of patients for inclusion into trials. On the back end of the trial, DL algorithms have the theoretical ability to predict patients who may respond to a new medicine based on the baseline images and clinical evaluation of the patient using the data sets generated in the clinical trials.

In clinical development, a key tenet is to get the right medicine to the right patient at the right time. AI and DL are capable of dramatically improving our ability to achieve this mission.