Table 2 Ablation study of MDOT components

From: Bridging clinical knowledge and AI: an interpretable transformer framework for ECG diagnosis

Method

Accuracy (%)

F1-Score (%)

Baseline (LeViT)

90.13

86.34

Baseline + Attention (A)

92.38

88.41

Baseline + A + OSC (B)

95.43

92.88

Baseline + A + B + Distillation (C, w/o XGBoost)

98.15

95.78

MDOT (Full Model)

99.53

97.26

  1. We report the mean accuracy and F1-score from incrementally adding the Attention module (A), OSC module (B), and Momentum Distillation (C). The final two rows isolate the contribution of the XGBoost component within the teacher model.
  2. Bold values indicate the best performance among all methods.
Back to article page