June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Optical coherence tomography segmentation of retinal fluids using deep learning
Author Affiliations & Notes
  • Huanxiang Lu
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Andreas Maunz
    F. Hoffmann-La Roche Ltd., Basel, Switzerland
  • Footnotes
    Commercial Relationships   Huanxiang Lu Roche, Code E (Employment), Roche, Code I (Personal Financial Interest); Andreas Maunz Roche, Code E (Employment), Roche, Code I (Personal Financial Interest)
  • Footnotes
    Support  F. Hoffmann-La Roche Ltd., Basel, Switzerland, provided support for the study and participated in the study design; conducting the study; and data collection, management, and interpretation
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 1124. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Huanxiang Lu, Andreas Maunz; Optical coherence tomography segmentation of retinal fluids using deep learning. Invest. Ophthalmol. Vis. Sci. 2023;64(8):1124.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : To build/validate artificial intelligence (AI)-based segmentation models for optical coherence tomography (OCT) volume scans; disease activity detection, quantitation, and monitoring; and precise machine learning/statistical modeling in neovascular age-related macular degeneration (nAMD).

Methods : Spectral domain OCT volumes from a clinical trial with 273 treatment-naïve eyes with nAMD at baseline, treated with ranibizumab 0.5 mg or various faricimab doses for 9 months, were selected post hoc and annotated by 2 experts from Liverpool Ophthalmology Reading Center on the B-scan level; each B-scan was annotated by a single grader. A sparse selection of 2–9 (average 6) B-scans per volume across 334 volumes, obtained from Spectralis (Heidelberg Engineering) OCT machines, were annotated by drawing contours of the intraretinal fluid (IRF; cystoid spaces), subretinal fluid (SRF), pigment epithelial detachment (PED), subretinal hyperreflective material (SHRM), and hyperreflective foci (HRF). A subselection of B-scan annotations was adjudicated by a senior clinician. Contours drawn on the B-scans were stored in raster format, and converted to label maps (fluids) and elevation maps (layers) of the original image dimension. A bug fix in the conversion software led to volume increases of SRF (+20%) and HRF (+70%) annotations. Data were split on the patient level into training (90%)/validation (10%) sets. UNet (convolutional neural network for biomedical image segmentation) was trained using pathologic features and layers in the annotated volumes (ie, pixel-level semantic segmentation). These were evaluated against validation set annotations using Sørensen–Dice coefficient (Dice) scores measuring overlap between annotation and model prediction.

Results : Median Dice scores were 85%/79%/79%/79%/65% for IRF/SRF/PED/SHRM/HRF. Compared with the previous version (no bug fix) of the data set (69%/73%/80%/65%/26%), there was a more homogeneous performance level, with improved SRF, SHRM, and HRF levels.

Conclusions : Precise, reliable OCT segmentation is important for monitoring/managing nAMD. Segmentation models using AI can fill the gap between human annotations (mostly qualitative) and quantitative assessment of pathologic markers or retinal thickness to enable granular disease characterization. Corrections in annotations of individual fluid types can greatly affect the performance of other fluids in a joint segmentation model.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×