Data documentation

The data files discussed here can be found in the example data folder of the GitHub repository, or in the Supplemental Data of the manuscript¹.

Images

The images provided with this work are composite images formed by combining many nuclei clumps from a full-slide image. The image names follow the following convention

(image|mask)_(clumpSize)[_normalized].tif

where

• The image files contain the actual images of the nuclei clumps without modification.
• The mask files contain the mask of each nuclei clump as obtained using an adaptive log-weighted Otsu thresholding on the full-slide image.
• clumpSize is one of the following values, 2, 3, 4, 5, or 67.
• The optional flag _normalized at the end gives the images of the nuclei clumps after intensity normalization. These are the images used with the previous methods we compared to in the manuscript.

Example:

image_2.tif

mask_2.tif

image_2_normalized.tif

Truth data

Each nucleus from the testImages has been labeled. The data is stored in the five matlab files

markedCenters_(clumpSize).mat

where clumpSize is the same as above. Each .mat file contains a single Nx3 array. Each row in the array corresponds to a labeled nuclei center. The first column gives the object number to which the nuclei belongs, the second column gives the y-coordinate (vertical axis), and the third column gives the x-coordinate (horizontal axis). The objects are numbered starting from in the upper left corner of the image and increasing down the columns. This is the same numbering as obtained using Matlab’s bwlabel(), or a similar function.

Here is an example of the object numbering and format of the truth data:

Obj #	y	x
1	27	18
1	23	32
2	79	19
2	71	31
…	…	…
8	81	120
8	69	130
9	127	132
9	123	118

5D data set

The 5-dimensional data set used in the manuscript is located in the Matlab file, damage_data_5D.mat. This file contains a single Nx5 array where each row is a new point and each column is a dimension. This data file is approximately 100 MB.