Implement a CLI for Nuclei Classification

DigitalSlideArchive / HistomicsTK

A Python toolkit for pathology image analysis algorithms.

https://digitalslidearchive.github.io/HistomicsTK/

Apache License 2.0

389 stars 114 forks source link

Implement a CLI for Nuclei Classification #285

Closed cdeepakroy closed 5 years ago

cdeepakroy commented 7 years ago

The inputs to this CLI should be:

Whole slide image path
Nuclei annotation file (*.anot)
Nuclei feature file (*.hdf5)
Takes a scikit-learn model file (*.pkl)

The outputs of this CLI should be:

Outputs an annotation file (*.anot) with +ve nuclei in one color and -ve nuclei in another color
Outputs probability map at low-res that will be rendered as a heat map

cooperlab commented 7 years ago

@neal-siekierski here is a link to the heatmap generation we created for cell classification with the active learning system: https://github.com/cooperlab/ActiveLearning/blob/8508d8cb938d5c39fa59ca612850e7a0adc0f36e/al_server/src/learner.cpp#L2434

A lot of trial and error to get the normalization and window sizes correct.

manthey commented 7 years ago

We have a heatmap view in geojs that scales a scalar value over arbitrary points in a field: http://opengeoscience.github.io/geojs/examples/heatmap/ -- see the earthquake data set for 1.3 million points. You can tweak all sorts of parameters live there. @cdeepakroy and I were just discussing adding something to the annotation spec to support this as a first-class output.

manthey commented 5 years ago

This was added in #337.