Skip to main content
. 2026 Jan 3;25(1):17. doi: 10.1007/s40200-025-01813-3

Table 6.

Ablation experiments

Configuration DSC (%) IoU (%) Accuracy (%) Remarks
Without HCAM 94.3 92.7 95.1 Missing hybrid attention causes poor localization
Without Capsule Layer 96.8 95.4 97.0 Feature relationships under-represented
Without ELM Classifier 97.2 96.1 97.8 Slight drop in convergence speed
Full Model (Proposed) 98.5 97.5 99.0 Best performance with full integration