cellphe 0.3.3

CellPhe: Toolkit for cell phenotyping from time-lapse videos

CellPhe

CellPhe provides functions to phenotype cells from time-lapse videos and accompanies the paper:
Wiggins, L., Lord, A., Murphy, K.L. et al.
The CellPhe toolkit for cell phenotyping using time-lapse imaging and pattern recognition.
Nat Commun 14, 1854 (2023).
https://doi.org/10.1038/s41467-023-37447-3

The Python package is a port of the original R implementation.

Installation

You can install the latest version of CellPhe from PyPi with:

pip install cellphe

Example

An example dataset to demonstrate CellPhe’s capabilities is hosted on Dryad in the archive example_data.zip and comprises 3 parts:

• The time-lapse stills as TIFF images (05062019_B3_3_imagedata)
• Existing pre-extracted features from PhaseFocus Livecyte. (05062019_B3_3_Phase-FullFeatureTable.csv)
• Region-of-interest (ROI) boundaries already demarked in ImageJ format (05062019_B3_3_Phase)

These should be extracted into a suitable location before proceeding with the rest of the tutorial.

The first step in the CellPhe workflow is to prepare a dataframe containing metadata identifying the tracked cells across all the frames, along with any pre-existing attributes. The segmenting and tracking can be performed within CellPhe, or pre-segmented and tracked data from two widely used software (PhaseFocus Livecyte & Trackmate) can be directly imported.

Segmenting and tracking

NB: This feature is still experimental, please report any bugs at the issue tracker

CellPhe provides 2 functions to segment and track an image sequence:

• segment_images: Segments images using Cellpose
• track_images: Uses the ImageJ plugin TrackMate to track cells between frames without requiring ImageJ to be installed

from cellphe import segment_images, track_images

segment_images takes 2 arguments: the path to the directory where the images are stored (where the folder 05062019_B3_3_imagedata was extracted to), and a path to an output folder where the resultant Cellpose masks will be saved.

This can take several minutes depending on the number of images and their resolution.

segment_images("05062019_B3_3_imagedata", "masks")

Confirm that the masks directory has been created and populated with TIFs containing cell masks. If it has, then you are ready to track the cells. track_images takes at minimum 3 arguments: the location of the masks created by segment_images, the filename to save the output metadata to, and a folder name to save the ROIs in. Optionally you can also save the ROIs as a zip so they can be easily opened in ImageJ, and change the tracking options - by default the Simple LAP method is employed.

track_images("masks", "tracked.csv", "rois")

Confirm that the tracked.csv file was created and the rois folder has been populated with ROI files. These outputs can now be loaded into CellPhe.

Importing pre-segmented and tracked data

Once a metadata file (CSV format) and a folder of ROIs are available, either directly output from external software (PhaseFocus Livecyte or TrackMate in ImageJ), or from within CellPhe as in the previous section, they can be read into CellPhe. The import_data function accepts metadata files from one of these sources and converts it into a standard format. It takes 3 arguments: the metadata file path, the source, and the minimum number of frames that a cell must be tracked for to be retained in the dataset (optional).

from cellphe import import_data

For example, the dataset that was segmented and tracked in the previous section can be imported as:

feature_table = import_data("tracked.csv", "Trackmate_auto")

Alternatively, the example below creates the metadata dataframe from the supplied PhaseFocus dataset, only including cells that were tracked for at least 50 frames.

input_feature_table = "05062019_B3_3_Phase-FullFeatureTable.csv"
feature_table = import_data(input_feature_table, "Phase", 50)

If a segmented and tracked dataset is available from a different source then it can still be used in CellPhe provided that it can be loaded into a pandas.DataFrame containing:

• Each row corresponding to a cell tracked in a specific frame
• A column FrameID (integer) denoted the frame number in chronological order
• A column CellID (integer) identifying the cell
• A column ROI_filename (string) denoting the filename (without extension) of the corresponding ROI file, not including the full path

Additional columns providing cell features can be included and will be retained and incorporated into the CellPhe analysis. The PhaseFocus dataset keeps the volume and sphericity features, for example.

Generating cell features

In addition to any pre-calculated features, the cell_features() function generates 74 descriptive features for each cell on every frame using the frame images and pre-generated cell boundaries, based on size, shape, texture, and the local cell density. The output is a dataframe comprising the FrameID, CellID, and ROI_filename columns from the feature table input, the 74 features as columns, and any additional features that may be present (such as from import_data()) in further columns.

from cellphe import cell_features

cell_features() takes as arguments the feature table, the folder where ROIs are saved, the folder where the images are, and the framerate. It expects frames to be named according to the scheme <experiment name>-<frameid>.tif, where <frameid> is a 4 digit zero-padded integer corresponding to the FrameID column, while ROI files are named according to the ROI_filename column.

roi_folder = "05062019_B3_3_Phase"
image_folder = "05062019_B3_3_imagedata"
new_features = cell_features(feature_table, roi_folder, image_folder, framerate=0.0028)

Generating time-series features

The next step is to calculate features that incorporate the time-dimension. This is done with the time_series_features function, which accepts a dataframe with the cell-level features as output earlier from cell_features.

from cellphe import time_series_features

Variables are calculated from the time series providing both summary statistics and indicators of time-series behaviour at different levels of detail obtained via wavelet analysis. 15 summary scores are calculated for each feature, in addition to the cell trajectory, thereby resulting in a default output of 1081 features (15x72 + 1). These are output in the form of a dataframe with the first column being the CellID used previously.

tsvariables = time_series_features(new_features)