Table 1 Ablation study analysis.

From: Multi-task deep learning framework combining CNN: vision transformers and PSO for accurate diabetic retinopathy diagnosis and lesion localization

Variant

Description

DR Classification Accuracy (%)

IoU Score (Localization)

Model 1–M1

Single view only (macula)

93.4

82.1

Model 2–M2

Dual-view, naïve concatenation (no cross-attention)

95.7

84.5

Model 3 - M3

Dual-view + Cross-attention (no PSO fusion)

97.2

86.4

Proposed Model–M4

Dual-view + Cross-attention + PSO-weighted fusion

98.9

88.7

Back to article page