Skip to main content

Table 1 Hyperparameter search space

From: Automatic segmentation of the great arteries for computational hemodynamic assessment

Parameter

Type

Domain

Scales

Architecture

{2, 3, 4}

Layers per block

Architecture

{2, 3, 4}

Initial filters

Architecture

{32, 64}

Learning rate

Training

[0.0001, 0.01]

Batch size

Training

{2, 4}

Loss function

Training

{CCE, Dice, IoU, Tversky, focal Tversky}

  1. { ∙ } represents a discrete set of values; [ ∙ ] represents a continuous interval of real values. The learning rate was sampled using a log-uniform distribution
  2. CCE categorical cross-entropy