- Tagline: Automated eye disease diagnosis
- Category: Featured Research
Context
Automating OCT analysis can improve speed and accessibility of retinal disease screening.
Role
Independent Researcher from model design through evaluation and presentation (IEEE AIC 2024).
Work completed
- Built Se-ResNet with Squeeze-and-Excitation channel attention
- Preprocessed 84,000 OCT scans with augmentation and contrast enhancement
- Trained with CUDA, LR scheduling, and early stopping
Results
- 96% classification accuracy
- 6% improvement over baseline ResNet
- Strong per-class precision and recall