• Fix and improve documentation
• Add an escape-hatch "other" modality
• Add peak types as modalities in KDE & WS

• Save and load fit artifacts for LSQ and Nyul for both the CLIs and Python API

• Major refactor to reduce redundancy, make code more usable outside of the CLIs, and generally improve code quality and documentation.

Fix find package in setup

fix name for pypi

Minor refactor to remove deprecation warnings, code cleanup, and publish to PyPI

intensity-normalization v1.3 release notes

Quality Metric

Added a quality metric for the normalization result (pairwise calculation of Jensen-Shannon Divergence of the histograms) and created a corresponding plotting routine to visualize the result.

RAVEL Functionality

Added functionality to increase robustness of the RAVEL result, specifically, added the --use-atropos flag to the executable script to enable the use of Atropos for the control (i.e., CSF) mask as described in the RAVEL paper. Default is FCM-based.

Nyul & Udupa Normalization

Changed the functionality of the hm-normalize function (corresponding to the Nyul and Udupa piecewise linear histogram matching routine) such that the transform is invertible, as described in the original paper. Previously the first and last percentile were saturated to the minimum and maximum value on the standard scale, but in this implementation the first and last percentile are extrapolated from the first and last linear fit (i.e., the first and last percentile are fit to the line determined in the 1%-10% interval and the 90%-99% interval). This fixes some odd behavior and better follows the original algorithm specifications.

Miscellaneous

• Fixed some documentation issues where the documentation did not reflect the correct script name
• Cleaned up interfaces
• Fixed bug in fcm-normalize single image option
• Changed behavior of gmm-normalize for better WM mean calculation, removed unused options

Initial production version. All methods have been tested.