Related Experiment Video
Updated: Aug 12, 2025
10:14
Author Spotlight: Ex Vivo OCT-Based Multimodal Imaging of Human Donor Eyes for Research into Age-Related Macular Degeneration
Published on: May 26, 2023
3.4K
Deep learning based diagnostic quality assessment of choroidal OCT features with expert-evaluated explainability.
S P Koidala1, S R Manne1, K Ozimba2
1Indian Institute of Technology Hyderabad, Kandi, 502284, India.
Scientific Reports
|January 28, 2023
Summary
Deep learning models accurately detect optical coherence tomography (OCT) image quality for choroidal analysis. EfficientNet-B3 demonstrated superior performance, aiding in standardized screening for eye diseases like AMD and CSCR.
Area of Science:
- Ophthalmology and Medical Imaging
- Artificial Intelligence in Healthcare
- Biomedical Engineering
Background:
- Vision-threatening eye diseases like age-related macular degeneration (AMD) and central serous chorioretinopathy (CSCR) stem from choroidal dysfunction.
- Current screening relies on optical coherence tomography (OCT) images, but algorithm performance hinges on scan quality.
- Standardized quantification tools require reliable methods for assessing choroidal feature quality in OCT scans.
Purpose of the Study:
- To develop and evaluate deep learning (DL) models for detecting the quality of choroidal features in Spectralis OCT images.
- To assess the attention of DL models on the choroid layer using explainability techniques.
- To correlate DL model attention with subjective grading scores for objective evaluation.
Main Methods:
- Trained three state-of-the-art DL models (ResNet18, EfficientNet-B0, EfficientNet-B3) on a dataset of 1593 good and 2581 bad quality Spectralis OCT images.
- Utilized color transparency maps (CTMs) based on GradCAM to visualize DL model attention on the choroid.
- Introduced overall choroid coverage (OCC) and choroid coverage in the visible region (CCVR) scores to objectively correlate visual explanations with model attention.
Main Results:
- The DL models achieved average accuracy and F-scores exceeding 96% in detecting OCT image quality.
- OCC and CCVR scores indicated that the DL models predominantly focused on the choroid layer.
- EfficientNet-B3 showed the strongest agreement with clinician assessments, outperforming other models.
Conclusions:
- The proposed DL framework accurately detects OCT image quality and demonstrates focused attention on the choroid layer.
- EfficientNet-B3 offers superior performance, making it suitable for benchmarking automated choroidal biomarker detection tools.
- The methodology can be adapted for quality assessment in other region-specific DL-based medical imaging tasks.
