Neurite

Includes unique layers like SpatiallySparse_Dense and LocallyConnected3D, not available in standard Keras, which address sparse data and local connectivity in 3D medical imaging.

Offers configurable models such as UNet and hourglass networks in tf/models.py, tailored for medical segmentation and analysis tasks with many adjustable parameters.

Provides generators in tf/generators.py for medical image volumes and segmentations, streamlining data loading and augmentation for complex 3D training datasets.

Includes metrics like Dice from tf/metrics.py that can be used as loss functions, along with utilities in tf/utils/seg.py for enhanced segmentation analysis and evaluation.