Related Experiment Video
Updated: Jun 18, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Evaluation of OCT biomarker changes in treatment-naive neovascular AMD using a deep semantic segmentation algorithm
Ben Asani1, Olle Holmberg2, Johannes B Schiefelbein3
1Department of Ophthalmology, Ludwig-Maximilians-University, Munich, Germany. ben.asani@med.uni-muenchen.de.
Eye (London, England)
|July 27, 2024
Summary
A novel deep learning algorithm accurately quantifies changes in OCT biomarkers for neovascular age-related macular degeneration (nAMD) patients undergoing anti-VEGF therapy. Significant fluid reduction occurs within 3 months, but RPE loss suggests potential long-term visual decline.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Neovascular age-related macular degeneration (nAMD) is a leading cause of vision loss.
- Anti-VEGF therapy is a standard treatment, but real-life quantitative changes in OCT biomarkers are not well-established.
- Accurate segmentation of OCT features is crucial for monitoring treatment response.
Purpose of the Study:
- To quantitatively assess real-life changes in OCT biomarkers in treatment-naive nAMD patients receiving anti-VEGF therapy.
- To develop and validate a deep learning-based semantic segmentation algorithm for analyzing 11 OCT features.
- To establish benchmark results for automatic segmentation of nAMD-related biomarkers.
Main Methods:
- A Deep U-net based semantic segmentation ensemble algorithm was trained for high-performance segmentation.
- The algorithm was applied to analyze Optical Coherence Tomography (OCT) features before, and at 3 and 12 months after anti-VEGF therapy.
- Performance was evaluated using F1 scores on a hold-out test set.
Main Results:
- The algorithm achieved high F1 scores (near 1.0) for neurosensory retina and subretinal fluid.
- Significant reductions were observed in intraretinal fluid, subretinal fluid, subretinal hyperreflective material, fibrovascular PED, and central retinal thickness after treatment.
- Intraretinal fluid, fibrovascular PED, and ERM predicted poor outcomes.
Conclusions:
- The developed segmentation algorithm enables efficient volumetric analysis of OCT scans for nAMD.
- Anti-VEGF therapy shows the most significant impact within the first 3 months.
- Gradual RPE loss indicates potential progressive visual acuity decline, necessitating further research into personalized therapy regimens.
