v0.14.1

Minor performance improvements

v0.13.0

What's Changed

• Add option to set force=True in dcmread
• Change how we handle creating temp files, should now properly be deleted after run
• V0.13.0 dev by @jsilter in #9

Full Changelog: v0.12.1...v0.13.0

Version v0.12.0

What's Changed

• Improve model downloading and storage. mirai-predict now has "--dry-run" option for testing and downloading of models.
• Remove unix-specific library dependency
• Check if dcmtk is installed, don't use it if not

Full Changelog: v0.11.0...v0.12.0

v0.11.0

Set num threads available to torch.

Available as command line argument, default is the number of CPUs. Doesn't matter when bare-metal but improves performance substantially in containers (assuming additional CPU cores present and available).

v0.9.0

• Add 'mirai-predict' CLI tool. This also means mirai can be installed and used with pipx.
• Rename "inference.py" to "predict.py", and move it to the onconet directory.
• Check that process_exam receives exactly 4 images, otherwise raise error. Previous behavior was still an error, but one which is harder to interpret.
• Modify installation dependencies. Don't install pandas or scipy just for inference, since they are not needed.
• Convert input images to grayscale. In most cases this will be a no-op, but apparently some images are RGB.

Full Changelog: v0.8.0...v0.9.0

v0.8.0

What's Changed

Refactor so that some dependencies can be made optional.

Main motivating factor was the calibrator needed scikit-learn, in particular a very old version. This has been causing issues. So we make a new calibrator class which mimics the inference behavior and can be pickled/unpickled with no dependencies (besides itself); the results are exactly the same. Tweak some other imports as well for similar purpose. Many files needed a bit of modification as a result.

• Modify setup.cfg so we don't need to install sklearn unless training.
• Leave the old config in for now, will remove it next version.
• Fix spelling of "calibrator"
• Add regression test over INbreast dataset. Does not run automatically.

Full Changelog: v0.7.0...v0.8.0

v0.7.0

What's Changed

• Individual scan inference by @jsilter in #3
• Get the log level from an environment variable by @jsilter in #4
  Full Changelog: v0.6.0...v0.7.0

v0.6.0

What's Changed

• Upgrade to use Python 3.8
• Add version numbers to dependencies
• Update README

This release includes example data and model weights as assets.