Mastodon numerical features.

This chapter describes how the feature values currently available in Mastodon are calculated. We also gives information about their dimension, units, etc.

Feature dimensions.

In Mastodon, feature values are expressed when possible in physical units. Each feature projection as a dimension (in the physics meaning) from which we compute the units of the values. For instance, if a feature value as the dimension LENGTH and the spatial units is μm, then the values of this feature will be in μm. Mastodon only has physical units for space. For time, the frame interval is always equal to the dummy 1 frame. This is why you will find all units involving time expressed in frames. Each feature projection has a dimension, and the features report what are the dimension of their projections in the feature computation dialog. The table below lists all the feature dimensions currently supported in Mastodon, and give examples of the derived units when the spatial units are μm.

Dimension

Name

Example units

Description

NONE

None

ø

Used for dimensionless quantities, such as frame position, number of things, etc

LENGTH

Length

µm

For quantities about the length of objects. For instance radius or distance between objects.

POSITION

Position

µm

Dimension for feature that report a position. Different from LENGTH so that for objects with small lengths at large positions, quantities are plotted separately.

TIME

Time

frame

For quantities that report a delay, a duration or the timing of an event. Because Mastodon does not deal with physical units for time, quantities formed with the time dimension always use the frame unit.

VELOCITY

Velocity

µm/frame

For quantities that report a speed or a velocity.

RATE

Rate

/frame

For quantities that report a change per units of time.

ANGLE

Angle

Radians

Measures of angles. In Mastodon, all angles are in radians.

STRING

NA

ø

For non-numeric features.

INTENSITY

Intensity

Counts

For quantities based on pixel values. For instance the mean intensity within a spot.

INTENSITY_SQUARED

Intensity²

Counts²

For quantities based on pixel intensity squared. For instance the variance of the mean within a spot.

QUALITY

Quality

ø

This dimension is used by spot detectors. There is a special feature called Detection quality, that stores for each spot they detect automatically a measure of quality or confidence in their detection.

COST

Cost

ø

This dimension is used by spot linking algorithms. There is a special feature called Link cost used in the estimation phase. It stores for each link the cost that the linker computes for it in the estimation phase. These costs are then used in the association phase to retrieve the best set of links.

Spot features.

Spot intensity.

This feature has six projections per channel:

  • Mean

  • Std

  • Min

  • Max

  • Median

  • Sum

These 6 projections are multiplied by the number of channels or sources in the image. You will find the projection names appended by ch 1, ch 2, etc, as exemplified below on an image with one channel.

The values are floating point numbers, with the dimension INTENSITY. These projections give the mean, max, min, … intensity at over all the pixels inside the spot ellipsoid. Like all features related to measuring pixel intensity, they take a long time to compute over large models.

Spot center intensity.

Computes the intensity at the center of spots by taking the mean of pixel intensity weigthted by a gaussian. The gaussian weights are centered on the spot, and have a sigma value equal to the minimal radius of the ellipsoid divided by 2. The image below illustrates how these weights look like inside spots and what pixels contribute the average reported by this feature.

Spot quick mean.

This feature compute the mean intensity of the pixel inside spots using the highest level in the scale pyramid (lowest resolution) to speedup calculation. It exists to offer a quiker way of reporting spot intensity for large models when multiple resolution level exist in the source image. It is recommended to use the ‘Spot intensity’ feature described above when the best accuracy is required

Other spot features.

Feature name

Projections

Description

Spot frame

idem

The spot frame.

Spot N links

idem

The total number of links, incoming and outgoing, of the spot.

Spot position

X & Y & Z

The spot center position, in physical units.

Spot radius

idem

The spot radius in physical units. For spots that are ellipsoids, returns a radius using the geometric mean of the spot ellipsoid radiuses. This approximation is such that the sphere with the reported radius and the spot ellipsoid have the same volume.

Spot track ID

idem

The ID of the track the spot belongs to. Track IDs are positive integer numbers starting from 0.

Track features.

’Track’ is the vocable we use in Mastodon for the weakly connected components of the graph. A track is made of all the links and spots that can be reached by jumping across links in any direction. In a lineage, a track corresponds to a single cell and all its daughters, grand-daughters, etc . Track features are value that are defined for a whole track. An example would be the number of spots in a track. In Mastodon, there is no special place to store track feature values. Track feature values are stored in the spots of the tracks, and listed in spot features. By convention, their name starts with Track and spot features starts with Spot.

Feature name

Projections

Description

Track N spots

idem

The number of spots in a track.

Branch-spot features.

Because branch-spots link to spots in the core graph, the branch-spot features all relate to the hierarchy or neighborhood in the branch-graph.

Feature name

Projections

Description

Branch N successors

idem

Reports the number of successors of a branch spot. That is: how many branches emerge from this branch-spot. The branch-spot of a cell that divides will have a value of 2 for this feature. The end of a track will have a value of 0. The beginning of a track, 1.

Branch depth

idem

Report the hierarchy level of a branch. The hierarchy of a branch is how many ancestors a branch has. For instance, the first branch of a track has a level of 0. After one cell division, the two daughter branches have a level of 1, etc. This feature value is used to build the hierarchy graph.