Table 1 Morphological and shape based metrics comparing a baseline generative VAE and diffusion model for unconditional cardiac generation

From: Probing the limits and capabilities of diffusion models for the anatomic editing of digital twins

Model

Morphological metrics

MMD ()

COV (%, )

1-NNA (%, )

 

Prec. ()

Rec. ()

FMD ()

CD

EMD

CD

EMD

CD

EMD

VAE

0.99

0.46

3.79

1.25

4.06

22.16

23.40

65.37

66.12

Diffusion (ours)

0.85

0.74

2.28

1.32

4.20

28.72

31.91

77.05

76.62

  1. Generative VAE’s suffer from reduced generation diversity in terms of both morphology and 3D shape, while diffusion models are able to sample a wide variety of cardiac segmentations at the cost of reduced fidelity. MMD-CD and MMD-EMD values were multiplied by 1000 and 100 respectively.