Discoveries from ARVO
Discoveries from ARVO
Source/Disclosures
Source:

Etheridge T, et al. A semi-automated machine-learning based workflow for ellipsoid zone analysis in eyes with macular edema: SCORE2 pilot study. PLoS One. 2020doi:10.1371/journal.pone.0232494.

Disclosures: Etheridge reports no relevant financial relationships.
July 09, 2020
2 min read
Save

Machine learning proves effective in measuring ellipsoid zones

Source/Disclosures
Source:

Etheridge T, et al. A semi-automated machine-learning based workflow for ellipsoid zone analysis in eyes with macular edema: SCORE2 pilot study. PLoS One. 2020doi:10.1371/journal.pone.0232494.

Disclosures: Etheridge reports no relevant financial relationships.
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.

Novel machine-learning using ellipsoid zone area defect revealed a correlation of improvement in mean visual acuity and decrease in defect, according to data presented at the Association for Research in Vision and Ophthalmology annual meeting.

Improvements in mean visual acuity letter score (VALS) were observed at months 1, 6 and 12, then decreased 44.4% at month 24 after no protocol-defined treatment for a year. The study comprised patients with macular edema due to central retinal or hemiretinal vein occlusion. Mean VALS was better in eyes without an ellipsoid zone (EZ) defect compared with eyes with an EZ defect at all time points.

The 2-year SCORE2 study assessed EZ on spectral domain optical coherence tomography (SD-OCT) and its association with VALS using a machine-learning workflow and SD-OCT volume scans at baseline. EZ assessment was not possible at baseline to due signal blockage in over 90% of eyes.

Researchers evaluated and segmented SD-OCT volume scans at baseline for months 1, 6, 12 and 24. Segmented layer coordinates were used to generate face thickness maps showing EZ defect, and the area within the central subfield (CSF) was measured using machine-learning. All SD-OCT scans received a qualitative assessment of EZ status as normal vs. abnormal by expert graders. Workflow created through open-source data analytics platform with the machine-learning methods showed reliable measurements of EZ area defect from en face thickness maps and proved comparable to traditional measurements.

At month 1, 48.7% of eyes had an EZ defect, which decreased to 39% at month 6 and 32.8% at month 12, then increased to 44.4% at month 24. Mean area of EZ defect within the CSF was 0.07 mm2 at month 1, 0.03 mm2 at month 6 and 0.06 mm2 at month 24. Mean VALS in eyes without an EZ defect compared to eyes with an EZ defect was better at all points in the study and area of EZ defect correlated negatively with VALS at month 1 (–0.41; P < .0001), month 6 (–0.36; P < .0001), month 12 (–0.35; P < .0001) and month 24 (–0.4; P < .0001). Machine-learning assessment of EZ defect with the reading center’s assessment of an abnormal EZ was strongly associated.

At all visits, VALS was better in eyes without an EZ defect compared with those including the defect, worsening at month 24 after not receiving protocol-defined treatment for 1 year.