lesion-metrics 0.1.6

metrics for evaluating lesion segmentations

lesion-metrics

Various metrics for evaluating lesion segmentations [1]

• Free software: Apache Software License 2.0

• Documentation: https://lesion-metrics.readthedocs.io.

Install

The easiest way to install the package is with:

pip install lesion-metrics

To install the dependencies of the CLI, install with:

pip install "lesion-metrics[cli]"

To install the dependencies for total lesion burden/volume computation (see lesion_metrics.volume), install with:

pip install "lesion-metrics[volume]"

You can also download the source and run:

python setup.py install

Basic Usage

You can generate a report of lesion metrics for a directory of predicted labels and truth labels with the CLI:

lesion-metrics -p predictions/ -t truth/ -o output.csv

Or you can import the metrics and run them on label images:

import nibabel as nib
from lesion_metrics.metrics import dice
pred = nib.load('pred_label.nii.gz').get_fdata()
truth = nib.load('truth_label.nii.gz').get_fdata()
dice_score = dice(pred, truth)

References

[1] Carass, Aaron, et al. “Longitudinal multiple sclerosis lesion segmentation: resource and challenge.” NeuroImage 148 (2017): 77-102.

History

0.1.6 (2021-08-16)

• Fix minor bug in IoU per lesion computation

0.1.5 (2021-08-16)

• Make package compliant with mypy

• Add lesion volume calculation

• Add support for IoU threshold for LTPR and LFPR

0.1.4 (2021-07-14)

• Fix type hints in volume correlation

• Correct name of LFPR (lesion false positive rate) to LFDR (lesion false discovery rate)

0.1.3 (2021-06-07)

• Fix correlation computation in CLI and other minor typos.

0.1.2 (2021-05-26)

• Update code style to black and improve docs.

0.1.1 (2021-05-14)

• Fix repo name.

0.1.0 (2021-05-14)

• First release on PyPI.