Page 9 - Live-cellanalysis handbook

From Images to Answers

Generating Actionable Data

After the pixels which satisfy all of the measurement criteria are the mask) or per object (e.g., area occupied by individual cells).
identified in an image, it is possible to operate on this binary Once again, the appropriate choice of labels, image processing,
mask of pixels. The mask may be analyzed whole (for total area, and object identification can require deep technical expertise,
or confluence measurements) or broken into multiple subparts, as the number of options available to differentiate objects is
for example when defining or counting objects in the image. very broad. For example, if you are looking for all red-labeled
Depending upon the labeling of the sample, e.g. label-free or nuclei that are also labeled with a green reagent (for example,
tagged with a specific marker such as a fluorescent reagent apoptotic cells labelled with Caspase Green), it is possible to
labeling a specific organelle or structure, a wide variety of identify individual cells first using a transmitted light image
statistics may be generated. In the case of fluorescent reagent- [mask 1], breaking that mask into objects representing cells
labeled images, these statistics may include the mean intensity using image processing tools like watershed split, and then
value of all the pixels in the mask, the total additive intensity, classifying those objects/cells based on the included red and
the minimum, maximum, or standard deviation of the collective green mean intensity of the included nuclei. This task is more
intensity, or the fluorescence mask may be used to count easily performed when the scientific question is well-defined,
numbers of objects. Statistics may be global for the image as the appropriate tools are utilized, and the images processed
just described (e.g. total size of the mask, or mean intensity of systematically, and without bias.

Analyzing Image Data at Throughput

Now that a specific set of operations has been constructed to different treatment groups and at different time points should
process and analyze a representative image, this same set of be expected. In analyzing a large image set, one must be assured
operations may be applied to all images in an experiment in that the set of operations is suitable across the data set (e.g.,
exactly the same manner. If this set of operations inadequately on dead or living cells). Traditional image analysis software
processes the population of images included in an experiment, it does not offer the ability to assess a variety of images in an
may be necessary to make adjustments to the set of processing efficient manner, and thus typical live-cell microplate assays
operations based upon the population of images collected can be unwieldy, at best. Software must address the needs of
for the task. In a live-cell imaging experiment performed in the researcher by performing all the steps required to convert
a 96-well plate, a data set containing thousands of images is images to data at the scale of long-term time-lapse experiments,
perfectly reasonable. Many data sets will be considerably larger and at the rate of acquisition, in order to best understand
when capturing multiple channels, e.g. red fluorescence/green biological processes while they are happening.
fluorescence/transmitted light, so variability between images in

7