CellMagicWand

Python implementation of ImageJ Cell Magic Wand tool originally created by Theo Walker and provided at https://www.maxplanckflorida.org/fitzpatricklab/software/cellMagicWand/.

This implementation uses the same dynamic programming and edge following algorithm as the original, but operates on Python Numpy arrays. This makes it easier to incorporate into automated image analysis pipelines. This implementation is more robust to seed point selection and can also optionally operate on 3D images using information from all z slices.

Usage

There are 3 primary functions in cell_magic_wand.py, all of which have the same calling conventions.

• cell_magic_wand_single_point(image, center, min_radius, max_radius, roughness, zoom_factor) is analogous to the original ImageJ implementation

• cell_magic_wand(image, center, min_radius, max_radius, roughness, zoom_factor, center_range) runs cell_magic_wand_single_point with 9 seed points located in a square around and including the given center. A point is included in the final ROI if it is in the majority of ROIs found from each of the 9 seed points.

• cell_magic_wand_3d(image, center, min_radius, max_radius, roughness, zoom_factor, center_range, z_step) runs cell_magic_wand on multiple z slices in a 3D image. A point is included in the final ROI if it is in the majority of ROIs found from the z slices. Note that this function can take considerably longer to run than above 2D functions depending on the z depth and the z_step argument.

Arguments

• image A 2D or 3D image in Numpy array format. A 3D image argument to cell_magic_wand_3D should have its z dimension as the 0th dimension. For TIFF images, import tifffile (or install using pip) and run image = tifffile.load([image_filename]) to get an image variable with the right format.

• center A seed point located as close to the center of the cell as possible. Should be a 2-element tuple of integers (y,x) (pixels down, pixels right). If you are having problems, make sure you have the right coordinate system.

• min_radius The minimum distance from the seed point that the edge detection algorithm will search. Increasing min_radius helps the tool avoid finding organelles and other boundaries inside the cell.

• max_radius The maximum distance from the seed point the edge detection algorithm will search. Decreasing max_radius helps keep the tool from circling multiple cells.

• roughness Controls the roughness of the edge of the detected ROI. Analogous to the roughness parameter of the ImageJ Cell Magic Wand tool but not exactly equivalent. The tool samples points at 2 * pi * max_radius * roughness angles around the center. Increasing roughness increases runtime but can result in better detection. The default value is 2 and values [1,10] are reasonable.

• zoom_factor Factor by which the image is upsampled prior to edge detection. In most cases, the default value of 1 is appropriate.

• center_range The distance to center of the additional 8 seed points used in the robust cell_magic_wand and cell_magic_wand_3d functions. The default 2, so for center (y,x), points (y-2,x), (y,x+2), etc. will be used. Set such that all points used will be inside your cells and near the center.

• z_step The z stride used in cell_magic_wand_3d. For an image with N z-slices, floor(N/z_stride) will be used.

Return Values

All cell_magic_wand functions return a binary mask that is the same size as the input image with 1s inside the detected cell and 0s elsewhere.

cell_magic_wand_single_point also returns an array of points that are on the edge of the detected cell

Tips

Even with multi-seed sampling, the center argument still considerably affects cell magic wand tool accuracy.

If your image has a lot of contrast variability, preprocessing it with a median filter can improve cell magic wand tool results. The easiest way is to use the median_filter function from scipy.ndimage.filters.