Automated, Machine Learning Algorithm Evaluates Retinal Layer Parameters in MFC

Automated machine learning algorithm automatically segmented optical coherence tomography (OCT) images of retinal layers in the setting of multifocal choroiditis (MFC), according to a study presented at the Association for Research in Vision and Ophthalmology (ARVO) Annual Meeting held in New Orleans, April 23-27, 2023.

Patients with MFC present with thickened outer retinal layers. However, correctly identifying such inflammation of the outer retinal layers can be difficult due to the structural disorganization of the retina.

Use of a machine learning algorithm may improve interpretation of OCT images in the setting of MFC. As such, this retrospective, observational study was performed by investigators at the Cleveland Clinic Cole Eye Institute. Women (N=5; n=8 eyes) with MFC who were receiving corticosteroid treatment were recruited for this study. Images from OCT obtained prior to and after initiating corticosteroid therapy were evaluated by a machine learning algorithm that automatically segmented retinal layers. Relevant MFC parameters were assessed.

The OCT images were obtained an average of 34 days apart.

At baseline, central subfield thickness (CST) was 263.89 (SD, 28.16) mm and ellipsoid zone (EZ) coverage at 0, 10, and 20 mm was 2.59% (SD, 3.64%), 3.27% (SD, 4.48%), and 3.80% (SD, 5.17%), respectively.

The change in CST at follow-up after initiating corticosteroid treatment was 0.75% (SD, 8.57%) and the change in EZ coverage at 0 mm was -41.23% (SD, 60.85%), at 10 mm was -43.56% (SD, 63.74%), and at 20 mm was -38.05% (SD, 64.11%).

The major limitation of this machine learning approach was that images of 7 eyes from 6 patients failed to segment and could not be included in this analysis.

“Machine learning analysis can be used to automatically segment retinal layers on OCT in patients with MFC and assess morphological changes following corticosteroid treatment,” according to the researchers. “Machine learning based-segmentation could be used to follow these complicated patients to allow earlier identification of active lesions on OCT. Machine learning analysis resulted in segmentation failure on some OCT scans which can be corrected by manual segmentation.”

Disclosure: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of authors’ disclosures.