Deep learning-based automated detection of retinal diseases using optical coherence tomography images
Problem
Automated detection of four retinal conditions (CNV, DME, DRUSEN, NORMAL) in OCT images for computer-aided diagnosis.
Why it matters
Retinal diseases are a significant diagnostic problem and early accurate detection affects patient outcomes.
What the AI did
An ensemble of four improved ResNet50-based classifiers automatically classified OCT images in a 4-class task using patient-level 10-fold cross-validation.
Results
0.973 accuracy (95% CI 0.971-0.975), 0.963 sensitivity, 0.985 specificity at B-scan level, matching or exceeding experienced ophthalmologists
Impact
Benchmark-level validation shows performance matching or exceeding experienced ophthalmologists; clinical deployment not stated.
Record
| Venue | Biomedical Optics Express |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Li Feng, Hua Chen, Zheng Liu, Xue‐dian Zhang, Minshan Jiang, Zhizheng Wu, Kai-qian Zhou |
| Citations | 193 (OpenAlex, at import time) |
| References | 42 |
| Model or method | Ensemble of improved ResNet50 instances |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1364/boe.10.006204 |
| Open access | gold |
| Keywords | OCT, retinal disease, ResNet50, deep learning, medical imaging |
Importance rating
real but narrower advance
Retinal disease classification called a significant CAD problem; strong benchmark results matching ophthalmologists, but no scale or clinical deployment stated.