Data dissemination: Reference Image Stitching Data

Stem cell colonies images using Phase Contrast and Cy5 fluorescent imaging modalities (60GB)

Day2 Plate – Overlaps 10-50% and Reference Colonies (18.8GB)
Day3 Plate – Overlaps 10-50% and Reference Colonies (20.1GB)
Day4 Plate – Overlaps 10-50% and Reference Colonies (21.2GB)


This dataset was acquired to test the accuracy of image stitching tools. It is acquired by imaging stem cell colonies seeded in one well of a 6-well plate using in phase contrast and fluorescent modalities. All plates are seeded at low density to allow for colony growth. They are then fixed at days 2, 3, and 4 respectively.

Reference imaging measurements

The exposure time of the Cy5 fluorescent imaging modality was selected to produce high contrast images where a manual segmentation threshold of 500 intensity unit (pixels of intensity greater than 500 are foreground) correctly segments an image. Each of the three reference datasets consist of a pool of centered reference colony images and their associated stage locations. The area and centroid locations are acquired after segmenting raw images with a global manually selected threshold and filtering small objects by imposing a minimum colony size of 3 stem cells (area > 2000 pixels). Each colony is centered inside the FOV by an automated feedback loop providing translations to geometric centroid of a thresholded segmented mask of that colony. The loop minimizes the difference between the colony centroid and the middle of the FOV and a convergence is reached when that difference is less than 1 μm. The final centroid location for a colony is saved from the microscope stage controller. The three reference datasets (Day2, Day3, and Day4) were acquired using an Olympus microscope controlled by MicroManager.

Dataset for testing stitching accuracy

For each well of the 3 plates, five sets of image tiles were acquired. Table 1 summarizes these datasets. Each set of image tiles covers the same inscribed square region on the plate as the reference measurements. This test dataset is imaged with overlaps ranging from 10 % to 50 %. The variation in the overlap between image tiles enables the evaluation of stitching algorithms with respect to the overlap.

Table 1: Reference Dataset Summary
Dataset acquisition Overlap area as FOV percentage Area covered (μm) Number of reference colonies
Day2 10,20,30,40,50 19500 x 16900 516
Day3 10,20,30,40,50 17300 x 17300 388
Day4 10,20,30,40,50 18300 x 17800 246