Segmentation identifies individual objects in the image. In the
segmented image objects are contiguous areas marked with unique
integer numbers (markers). Image Analyst MKII can segment binarized
images, grayscale images, grayscale images with uneven background
using local thresholding and grayscale images by extending
previously calculated seeds.
Built in pipelines use various combinations of the
Advanced Segmentation functions detailed below e.g.:
Hint: To see worked examples of preparation of images for
segmentation and multi-step segmentations with classifiers, activate
these pipelines and look the pipeline structure by pressing the
Pipeline Window
main toolbar window.
To segment an image (series) in Image Analyst MKII
- First binarize images with the
Threshold
function, then use the
Simple
Segmentation function (both are accessed from the main
menu/Segmentation)
- Or use the
Advanced
Segmentation to segment grayscale images with the Watershed
algorithm. Advanced segmentation is a combination of
locally adaptive thresholding and Watershed segmentation.
- To calculate segments originating from specific seeds, e.g.
to find boundaries of cells originating from nuclei, use the
Advanced
Segmentation Separate Seed function. The seed
can be another grayscale image, or a previously segmented image.
Using Simple
Segmentation
-
Prepare image for thresholding
- Binarize image with the Segmentation/
Threshold
function
- Set segmentation classifiers (see
below)
- Segment image with the Segmentation/Simple
Segmentation
- Textual output: to see
morphological parameters of the objects set it Yes.
If segmenting for further processing No
prevents opening of the data window. If Excel Data Window is
open, results will appear there.
Using Advanced Segmentation
-
Prepare image for segmentation
(do not binarize)
- Set segmentation classifiers (see
below)
- Segment image with the Segmentation/Advanced
Segmentation
Parameters of this function closely follow the parametering
of the
Threshold
function in its
locally
adaptive mode.
- Way: Above or
Below for detection of brighter or dimmer
details than their surroundings, respectively. The
parametering is described for brighter details below:
- The Threshold value calculation method
is:
- Pixel Value: local maxima
brighter than their surroundings by this value will be
detected
- Percentile by frames: local
maxima brighter than their surroundings by the level calculated as the given percentile of the
intensity histogram of each image will be detected
- Percentile by series: local
maxima brighter than their surroundings by the level calculated as the given percentile
of the intensity histogram of the image series will be
detected
- Otsu by frames: local maxima
brighter than their surroundings by the level calculated as the
Otsu optimal threshold of each image will be detected
- Otsu by series: local maxima
brighter than their surroundings by the level calculated as
the Otsu optimal threshold of the image series will be
detected
- The Threshold from local max/min
is:
- None: no threshold will be applied,
the segments will completely fill the image
- Bound uniformly: the boundary of objects
is defined by that they are brighter than their
surroundings by the above calculated threshold value.
Practically the threshold value is applied downwards
from the local maximum. The Determine
boundaries at is normally set to zero.
This value is added the threshold value when determining
the boundaries of an object, so increase this value to
get thicker objects.
- Bound locally: the boundary of the object will be determined at the
percent of its maximal intensity, given in the
Determine boundaries at parameter.
- Method:
- Mark Seed: pixels corresponding to local maxima are
shown
- Mark Shape: segments to contiguous objects without
applying Watershed algorithm
- Watershed: splits contiguous objects using local
maxima as sources in the Watershed algorithm
- Connectivity: defines contiguity
- Separate segments: if Yes, touching
segments are separated by a single pixel line of zero
pixels.
- Evaluate Segment data
and Textual output: to see
morphological parameters of the objects set it Yes.
If segmenting for further processing No
prevents opening of the data window. If Excel Data Window is
open, results will appear there.
Using Advanced Segmentation with a separate seed
image
- Segment seed image as given for "Using Advanced
Segmentation" above. Note, that for using nuclei as seeds it is
useful to filter images before segmentation with the Filters/Detect
Nuclei Convolution to prevent splitting nuclei into multiple
segments.
- Segment the image (Image A)
where the seeds will be extended using the Segmentation/Advanced
Segmentation Separate Seed, setting the above generated
seed image as "Image B"
- The Threshold value calculation method
is:
- Mean of Marks: The mean intensity
measured in Image A in pixels corresponding to the seed
of the segment from Image B is used for calculating the
boundaries of the segment according to the "Threshold from local max/min"
parameter - see above at "Using Advanced Segmentation".
- Min of Marks: same as above, but
with the minimum of intensities in Image A pixels
corresponding to the seed.
- Max of Marks: same as above, but
with the maximum of intensities in Image A pixels
corresponding to the seed.
Morphological parameters measured by the Segmentation
functions:
Note that segmentation acts independently in each frame of the
time-lapse recording, thus in 2D. Some of the parameters below have
been designed for future compatibility with 3D segmentation.
- x,y : position of the center of the segment
- frame : time point or z position of the
center of the segment
- length : length of the segment in the third
dimension (z or t) – note that segmentation is currently 2D
only, so this is 1
- V: volume as total number of pixels of the
object (sum of surfaces if present in
multiple planes). For single frame objects A=V=S (note that this
is always the case as segmentation is currently 2D only)
- A: surface as total number of pixels of the
object in the plane (maximum if present in multiple planes;
=A=V for 2D)
- P: perimeter as number of pixels along the
perimeter of the object (maximum if present in multiple
planes)
- d: diameter(maximum if present in multiple
planes); the diameter is the diagonal of the bounding rectangle.
- S: surface (sum of perimeters and bounding
surfaces if present in multiple planes)
- ShF:
shape factor or circularity (mean if present in multiple
planes) = 4*Pi*A/P^2 - note that this value may be larger than 1
for small objects because P is the number of pixels along the
perimeter and not the real perimeter.
- ShF3D: spherical shape factor =
36*Pi*V^2/S^3
- Brnch: number branch points (meaningful
only for previously skeletonized objects)
- Fiber length: an approximation of length of
elongated objects based on surface area and perimeter = (P+Sqrt(P^2-16*A))/4.
- Fiber breadth: an approximation of width of
elongated objects based on surface area and perimeter = (P-Sqrt(P^2-16*A))/4.
Using Segmentation Classifiers
Object can be discarded from the segmentation using segmentation
classifiers. Objects that do not pass the classifiers are deleted
from the image (are filled up with zeros). Segments can be
classified based on their shape parameters or based on intensities
measured in other channels:
- Before segmentation use the Segmentation /Set
Segmentation Classifiers function running it twice; once
with Max and once with Min
- Limit Type: Max
: objects which have a any parameter that is greater than
the corresponding set value are discarded. Zero values are
not checked.
- Limit Type: Min
: objects which have a any parameter that is smaller than
the corresponding set value are discarded. Zero values are
not checked.
- Segments can be also classified based on
intensities measured in corresponding pixels of other channels using the
Segmentation /Set
Segmentation Intensity Classifiers function. Run the
function with Max or with Min limit type, and set from which
channel (Image Window) to get classifying intensity values as
Image B.
- Limit Type: Max
: objects which have a any parameter that is greater than
the corresponding set value are discarded. Zero values are
not checked.
- Limit Type: Min
: objects which have a any parameter that is smaller than
the corresponding set value are discarded. Zero values are
not checked.
Working with segmented images
- To measure number of objects or size and shape parameters of
each object or statistically (use the textual output of
segmentation functions, or read lists of values from Mathematica)
- To measure intensities of other images over the pixels
corresponding to the objects obtained by segmentation
- To count colocalization of different markers with the
objects
- To binarize images in a controlled manner,
allowing/rejecting objects based on size or morphological
criteria
- Use the
Threshold
with Pixel value=1, Above, None settings to re-binarize
segmented image.
- To find instances of the same object in consecutive frames
use tracking:
Limitations
The segmentation algorithm is not limited to maximum 65534
segments any more.
To increase speed, real diameters of objects are not calculated, the
diagonal of the bounding rectangle is used instead.
The perimeter is given in numbers of pixels along the perimeter,
that can be less than the actual distance around the perimeter, and
this makes shape factors the have values above 1 for small circular
objects.
The segmentation is 2D currently, thus each frame is independently
segmented.
Updated: 11/24/2015
Workflow
Track segments with one intensity channel
