Table 14 Feature importance comparison using SHAP and LIME for DR detection on retinal fundus images dataset.
From: An enhanced diabetic retinopathy detection approach using optimized deep learning technique
Feature | Feature Index (DGOA) | Importance Score (SHAP) | Importance Score (LIME) |
|---|---|---|---|
Microaneurysm Density | F₁₅ | 0.35 | 0.33 |
Exudate Presence | F₂₄ | 0.30 | 0.28 |
Hemorrhage Spread | F₁₀ | 0.25 | 0.22 |
Retinal Vessel Width Pattern | F₃₅ | 0.20 | 0.18 |
Macular Edema Indicator | F₄₂ | 0.18 | 0.16 |
Foveal Area Contrast | F₅₈ | 0.15 | 0.12 |
Vascular Tortuosity | F₆₉ | 0.12 | 0.10 |
Lesion Count Surrogate | F₄₇ | 0.10 | 0.08 |
Vessel Density Approximation | F₈₄ | 0.08 | 0.06 |
Retinal Texture Disruption | F₉₇ | 0.06 | 0.05 |
