EfficientNet variants evaluated for retinal disease classification from OCT images
Original title: Within-Family EfficientNet Scaling under Fixed-Resolution Transfer Learning for Spectral-Domain Optical Coherence Tomography Retinal Disease Classification
B1 EfficientNet achieves highest accuracy for four-class SD-OCT classification, outperforming larger variants.
The study evaluates eight ImageNet-pretrained EfficientNet variants from B0 to B7 for four-class classification of retinal diseases from spectral-domain OCT images on the Kermany dataset. B1 achieves 96.95% test accuracy and highest Drusen F1 score of 0.8780 while larger models do not consistently improve performance. Accuracy-to-parameter ratio falls from 19.82 to 1.48 percentage points per million parameters. Results indicate task and protocol dependent scaling effects.
- Test accuracy varies non-monotonically from 95.93% (B0) to 96.95% (B1)
- B1 has highest Drusen F1 score of 0.8780
- Accuracy ranking differs from ImageNet hierarchy
- Final train-test gap weakly associated with test accuracy (r=0.26, p=0.53)
- AI models
- EfficientNet
From the original article
1 min read at arxiv.orgBackground/Objectives: CNNs perform well in retinal disease classification from spectral-domain optical coherence tomography (SD-OCT), but whether EfficientNet's natural-image scaling hierarchy transfers to medical-image transfer learning at fixed resolution remains unclear. This study evaluates EfficientNet-B0-B7 for four-class SD-OCT classification and accuracy-efficiency trade-offs.
By Shkelqim Sherifi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.