Eric W Schneider, MD: Comparing AI-Based Home OCT to In-Office OCT Scans

A comparison of artificial intelligence (AI)-based home optical coherence tomography (OCT) to in-office OCT suggested its effectiveness in imaging patients with fluid from retinal diseases, according to new data presented at the 128th Annual American Academy of Ophthalmology (AAO) Meeting.

In an interview with HCPLive, presenter Eric W. Schneider, MD, Tennesse Retina, described the system of home OCT, encompassing a small, form-factor SD-OCT device to acquire the images and an AI-based algorithm to review and segment images based on fluid volumes. Using the estimates generated by the AI-based algorithm, clinicians can identify a trajectory and find a way to monitor patients with retinal diseases efficiently.

“I really want to emphasize that it is more of a system than a single device. Either alone isn’t nearly as helpful as it would be paired together,” Schneider told HCPLive.

The first part of the pivotal trial, VISUALIZE, compared the home OCT device to in-office OCT for qualitative identification of hyperreflective spaces in OCT volumes. Patients were imaged for 5 weeks, with mandatory study visits—on the day of an in-office visit, a comparison was made between the home OCT scan and the in-office scan to see how closely the devices mimicked one another.

This comparison revealed “excellent agreement,” according to Schneider, as the positive percent agreement was 86% and the negative percent agreement was 87%. A pre-specified endpoint was met, defining good agreement as ≥80% for both positive and negative agreement.

These data also demonstrated the capability of patients to efficiently use the home OCT device, with 96.1% of individuals able to calibrate, activate, and perform the self-imaging at home. Mean scans per week were 5.9, with a mean time per scan of 48 seconds. According to Schneider, patients could not only effectively use the home OCT, but consistently used the device.

The second part of the pivotal trial evaluated the ability of the AI algorithm to quantify hyperreflective space volumes. Patients were imaged multiple times on separate home OCT devices in the office and then an in-office OCT to quantify repeatability and agreement.

Schneider suggested the importance of repeatability in the output of the home OCT and algorithm, as otherwise, the volume trajectories will be inconsistent. Multiple home OCT scans were evaluated by the AI grading output, compared with a single in-office OCT scan assessed by 3 independent reading center graders.

The coefficient of variation for patients with ≥10 volume units of total hyporeflective spaces was 11.1% for the AI-based home OCT and 16.4% for the in-office OCT and reading center graders, indicating it is a repeatable device.

For agreement, reading center graders graded the home OCT volume scans and compared them with the AI algorithm for the same scans, as well as between graders. These data showed equivalency in the Dice Similarity score, showing the volume estimation is similar for both the AI algorithm and the human graders.

“I think the study is somewhat comprehensive, but showing that both components of the device are able to work effectively to image patients and produce an output that's usable in a clinical setting is notable,” Schneider added.

Disclosures: Relevant disclosures for Schneider include Notal Vision and Carl Zeiss Meditec, among others.

References

_{Schneider EW. Pivotal Trial Outcomes of AI-Based Home OCT: A Qualitative and Quantitative Comparison with In-office OCT. Presented at the American Academy of Ophthalmology (AAO) 2024 Meeting. Chicago, Illinois. October 18-21, 2024.}