v1.1.1.post3

• Countplot available in plotting functions associated to the table widget
• New function to merge several binary classes (0 or 1) into a categorical class (0,1,2,3...)
• Write the image names from the train and validation set in the input_config.json file of a newly trained segmentation model
• Fix bug of "scale_model" missing when training a StarDist model
• Fix bug in the exclusion of NaN when optimizing the values of the background image to correct for the background using the model-free approach

See the full Changelog: v1.1.1.post1...v1.1.1.post3

v1.1.1.post1

Minor increment for stability fixes and small features.

• Documentation: preprocessing functions
• Table widget: new option to plot measurement distributions directly with tracked data, without requiring to collapse the time axis
• Segmentation:
  2D interpolation of NaNs in images before segmentation
  Fix rescaling to reach a cell size of 30 px not applied when predicting with a Cellpose model
• Filtering:
  Interpolate image before filtering to avoid the propagation of NaN values
• Threshold configuration wizard:
  A new option to bypass marker detection and watershed, and instead label all non-touching objects post threshold
• Classifier widget:
  fix a bug on the R2 computation
  increased number of iterations for the sigmoid fit

Full Changelog: v1.1.1...v1.1.1.post1

v1.1.1

• Bug fixes for v1.1.0

Full Changelog: v1.1.0...v1.1.1

v1.1.0

What's new

Preprocessing

We introduce a new preprocessing module, to correct the background of image stacks before segmentation. The user can choose among two background correction methods.

• The first is to fit a 2D model (paraboloid, plane) to each image of the stacks where the non-homogeneous part is filtered out using a threshold on the STD-filtered image. This preprocessing was previously available in the measurement options, but the user could not export the preprocessed image. Both options are now possible.

• The second is a model-free estimate method for the background using the multi-position information to reconstruct a median background within a well. This is a typical background reconstruction method for interference microscopy images (IRM, RICM). The user can select an optimal frame range where cells are less likely to be visible, threshold the least homogeneous parts of the image with the STD-filter technique, and visualize the background reconstructed for each well. If the background is satisfying, it is applied to each image, with an automatic regression to accommodate slight focus loss and intensity variations within a well.

Preprocessing protocols can be combined to preprocess independent channels differently. The output is a new stack with the "Corrected" prefix, which can be set as the stack of interest when performing segmentation and measurements.

Threshold-based cell classification

The classifier widget has been expanded with more options to classify tracked cells. The user can now classify cells according to their most likely state (median state), assuming no state change. This option is complementary to the previously implemented "irreversible event" that was able to describe a transition between two states.

Furthermore, you can now exclude a cell class from survival analysis.

Neighborhood

We introduce a new and intuitive neighborhood method that matches touching masks as neighbors. To accommodate ambiguities in cell-cell contacts, this type of neighborhood has a tolerance parameter: the border size dilation applied to the masks before neighborhood computation. The counting methods remain strictly identical to the isotropic distance method, introduced previously.

Log

Experimental: a logger is introduced with this version. A json is written in each position folder to keep track of how a position is processed.

Viewers

A viewer class is introduced to gradually replace all image viewers in Celldetective (except for napari). Several new viewers have been introduced to facilitate the setting of analysis parameters:

• a viewer next to the position list, in the control panel header, enables to preview the current stack
• several viewers help set threshold parameters a view preprocessing steps
• a viewer was augmented to set the contour distance in the measurement options
• a viewer was augmented to represent a circle of radius controlled with a slider to set an isotropic neighborhood or estimate cell size when calling a pre-trained Cellpose model

The viewers are accompanied by a scalebar either in µm or pixel depending on the purpose of the viewer.

Maintenance

• unitary tests for a large part of the Celldetective core
• code refactoring and layout subclasses
• widget behavior fixes, less crashes
• introduction of searchable combo boxes from superqt and colormap combo boxes
• bug fixes

Contributors

• @ksenijader
• @remyeltorro

Full Changelog: v1.0.2...v1.1.0

v1.0.2

First version available on PyPi.

What's new

• Annotation module for static data with measurement visualization/phenotype annotation within characteristic groups

• New measurements: intensity spatial distribution, spot detection
• Background correction before measurements for individual channels (both local to the cells and over the field)

• Classification tool compatible with static and dynamic data
• New signal data augmentation to create more left-censored events
• Bug fixes

Contributors

• @ksenijader (main)
• @remyeltorro

Full Changelog: v1.0.1...v1.0.2

v1.0.1

Stable release for Celldetective

• Not available on Pypi yet
• Still closed to the public

What's new

• Automatic estimation of event time when using the feature-classification tool, by fitting a sigmoid over the binary status signal
• Partial annotation of images for segmentation using napari ROIs
• Automatic correction of segmentation annotation errors (same labels for several cells)
• New automatic app resizing when collapsing the process blocks
• Fix multi-threading for segmentation
• Keep GPU / multi-threading options on new Celldetective sessions
• Bug fixes

v1.0.0

first stable release for Celldetective

• Not available on pypi yet
• still closed to the public