Performance Metrics for 5 Retinal Pathologies from RFMiD Dataset

LiteFeatNet, a lightweight CNN, achieved a testing accuracy of 90.33% on retinal disease classification. The model was trained and evaluated on 1,824 images from the Retinal Fundus Multi-Disease Image Dataset (RFMiD). Author Usman Rafi published these performance metrics on figshare in May 2026.

Use Cases

• Benchmark lightweight CNN architectures based on reported accuracy, precision, recall, and F1-score metrics.
• Compare model inference speed based on the reported 4 milliseconds per image processing time.
• Evaluate model generalizability using the results from the external RFMiD 2.0 dataset mentioned in the description.
• Assess computational efficiency based on the model's compact size of 19.87 MB and reduced parameter count.

Strengths

• Performance metrics are directly comparable, with LiteFeatNet outperforming 12 state-of-the-art models.
• Results are validated through an external dataset (RFMiD 2.0) and an ablation study.
• Dataset is small (5.5 KB), making it quick to download and analyze.

Limitations

• The dataset contains only aggregated performance metrics, not the underlying image data or model predictions.
• Row count is unknown, which may limit suitability assessment for detailed statistical analysis.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

figshare

Collection Method

Results from experiments training a CNN (LiteFeatNet) on the RFMiD dataset.

Freshness

Last updated 2026-05-11 17:32:54; freshness should be verified.