Image-Driven Multimodal Deep Learning for Skin Cancer Diagnosis: Cross-Attention Fusion of Dermoscopic Imaging Clinical Data and Knowledge Graphs

. 2026 Jul 25;26(15):4728. doi: 10.3390/s26154728

Algorithm 6 Bootstrap Statistical Significance Testing

1:
procedure Bootstrap( $n_{i t e r} = 10000$ , $m o d e l_{A}$ , $m o d e l_{B}$ , $t e s t_s e t$ )
2:
$Δ_{a c c} \leftarrow []$
3:
for $i = 1$ to $n_{i t e r}$ do
4:
$s a m p l e \leftarrow resample (t e s t_s e t, with_replacement = T r u e)$
5:
$a c c_{A} \leftarrow evaluate (m o d e l_{A}, s a m p l e)$
6:
$a c c_{B} \leftarrow evaluate (m o d e l_{B}, s a m p l e)$
7:
$Δ_{a c c} . a p p e n d (a c c_{A} - a c c_{B})$
8:
end for
9:
$μ_{Δ} \leftarrow mean (Δ_{a c c})$
10:
$σ_{Δ} \leftarrow std (Δ_{a c c})$
11:
$C I_{95} \leftarrow [μ_{Δ} - 1.96 σ_{Δ}, μ_{Δ} + 1.96 σ_{Δ}]$
12:
$p_{v a l u e} \leftarrow \frac{2 \cdot min (\sum 1 (Δ_{a c c} > 0), \sum 1 (Δ_{a c c} < 0))}{n_{i t e r}}$
13:
return $p_{v a l u e}, C I_{95}$
14:
end procedure