Algorithm Description Guidelines

Algorithm Description Guidelines In addition to submitting a docker container, participants must submit by August 12th 2024 a short description of their algorithm highlighting the main features.

Use the following form: FeTA Challenge Algorithm Description Submission

In addition to the above form, please provide a description of your model highlighting the main features. This description must include the following details (unless the parameter is not applicable for your model):

  • Model architecture

  • Number of layers

  • Convolution kernel size

  • Initialization

  • Optimizer

  • Cross-validation used?

  • Number of epochs

  • Number of trainable parameters

  • Learning Rate and schedule

  • Loss Function

  • Dimensionality of input/output (ie: 2D,3D, 2D+, etc.)

  • Batch Size

  • Preprocessing steps used (ie data normalization, creation of patches, etc.)

  • Data Augmentation steps (ie – rotation, flipping, scaling, blur, noise, etc.)

  • External dataset used? (allowed, but it needs to be publicly available

  • Framework (ie – MONAI, nnUNet, etc.)

  • Number of models trained for final submission

  • Post-Processing Steps (ie – ensemble network, voting, label fusion)

  • Clearly state which aspects are original work (if any) or already existing work

  • Include relevant citations, as well as if existing code/software libraries/packages were used

  • Which FeTA cases were included in the training and testing (ie – all cases, only pathological, only 1 institution, etc.)

  • Training/validation/testing data splits

  • Hyperparameter tuning performed

  • Training time

Note: If your method was not deep learning-based, please provide the equivalent appropriate information as listed above.