MIScnn Versions Save

A framework for Medical Image Segmentation with Convolutional Neural Networks and Deep Learning

v1.4

2 years ago

MIScnn release version 1.4

Changelog:

  • Started working on foundations for multi-model support (experimental)
  • Started working on a dynamic visualizer (work in progress)
  • Added smoothing epsilon to Normalization Subfunction
  • Added subfunction support for activation_output=True predictions
  • Improved Resampling & resize performance
  • Some overall quality of life improvements for activation output of predictions
  • Fixed learning_rate typo in Neural_Network class
  • Added Dictionary_IO support for passing pixel spacing (for resampling)

v1.3

2 years ago

MIScnn release version 1.3

Changelog:

  • Added multiple loss functions from Yeung et al.: Combo loss, Focal Tversky loss (symmetric and asymmetric), Focal loss (symmetric and asymmetric), Unified Focal loss (symmetric and asymmetric)
  • Updated dependencies as well as fixed several issues with new module versions
  • Updated to Tensorflow 2.7.0
  • Added a few more examples (RetinalSeg, BraTS2020)
  • Added new architectures: Attention- and Attention-Residual U-Net

v1.2

2 years ago

MIScnn release version 1.2

Changelog:

  • Fixed Python support only for 3.7 & 3.8 (CI/CD)
  • Updated various dependencies (Tensorflow 2.5.0)
  • Dropped Travis-CI support and integrated GitHub Actions
  • Fixed a grayscale Image IO interface bug
  • Enhanced plain U-Net architecture
  • Implemented CLI interface for MIScnn
  • Reworked detail storage of samples in MIScnn
  • Enhanced cross-validation functionality
  • Enhanced unittesting scenarios
  • Fixed a bug in the DataGenerator
  • Various quality of life improvements

v1.1

3 years ago

MIScnn release version 1.1.

Changelog:

  • Added Multi-threaded Subfunction preprocessing
  • Added absolute transformation of voxel spacing in DICOM interface
  • Added function for all sample loading
  • Added array type transformation for preprocessed Subfunction disk storage
  • Added unittesting for new features & functions
  • Added tutorial for custom architectures
  • Changed multi_gpu from Keras to Tensorflow MirroredStrategy
  • Changed parameter of Neural Network class 'gpu_number' (integer) to 'multi_gpu' (boolean)
  • Updated Tensorflow version from 2.3.0 to 2.3.1
  • Updated other module dependencies
  • Fixed some bugs

v1.0

3 years ago

MIScnn release version 1.0.