Unmix Absorbances

Separate an RGB image into two absorbance components. This command is designed for quantification immunohistochemical stains. For example, diaminobenzidine pigment (brown) can be isolated from a haematoxylin (blue) counter-stain. The image is log-transformed to recover absorbance values on each channel, than a linear unmixing algorithm is used to apportion absorbance between the two components. Three images are returned, maps of each absorbance component and a normalized residual map. The performance of this command depends on accurate specification of pigment values for each of the two absorbance components.

Input Parameters

Images

Arbitrary bit-depth RGB image.

Type: Image File (net.imagej.Dataset)

Pigment colors (R G B)

Color specifications for each of the absorbance components, formatted as text strings of three numbers, separated by spaces, in red, green, blue order. These parameters represent colors resulting from an intermediate amount of each pigment. Best performance is achieved with color specifications in the middle of the image dynamic range (for example, 150 60 60 for a red-brown pigment in an 8 bits per channel image).

Type: Text (String)

Results

Absorbances

Intensity maps of each absorbance component (a1 and a2).

For each pixel, minimizes the squared residual of yrgb = Pa12

yrgb, the RGB absorbance vector, where absorbance is calculated as -log(value/bit-depth) .
P, a matrix specifying pigment absorbances on the red, green, and blue channels
a12, the absorbance intensity of each component

The returned images are 16-bit.

Type: Image File (net.imagej.Dataset)

Residual

Normalized residual map, scaled to fit within the image dynamic range.

Type: Image File (net.imagej.Dataset)

Reference

Taylor CR and Levenson RM. Quantification of immunohistochemistry—issues concerning methods, utility and semiquantitative assessment II. Histopathology. 2006. doi: 10.1111/j.1365-2559.2006.02513.x