Hi Scott

Will not black do what you want?

R=G=B=0 results in no change to the underlying pixel except scaling
of the luminance by the relative opacity.

See the equation in PS 3.4 N.2.4.3.

David

PS. I presume you are storing your CAD array of probability of malignancy
values in a Segmentation Storage SOP Instance ? You should be, since that
is what it is intended for.

Scott Thompson wrote:
> Suppose a CAD algorithm analyzes contrast enhanced MRI and creates a
> value at each pixel in the range [0-25]; the higher the value, the
> more suspicious the pixel is for cancer.
>
> Blending Softcopy Presentation State (BSPS) has the ability to present
> fused displays consisting of the original grayscale underlying image
> and the CAD results as a pseudo color overlay image. This sounds like
> what we want.
>
> The pallets color LUT Module within BSPS describes the color mapping
> to be used for the superimposed image set.
>
> Suppose we want to start the mapping at index "7", e.g.,
>
> Index Color
> -------------------
> 7-10 YELLOW
> 11-25 RED
>
> The standard allows us to set the first pixel value mapped ("7" in
> this case). However, it says (PS 3.3 2008, C.7.6.3.1.5):
>
> "All image pixel values less than the first value mapped are also
> mapped to the first entry in the LUT..."
>
> Hmmm. I interpret that to mean that pixel values [0-6] would then be
> mapped to YELLOW, which is not what we want. Moreover, since all
> pixels have values in the range [0-25], it follows that every pixel
> should be mapped to some color in the LUT.
>
> Seems what I want instead is to start the mapping at index "0", e.g.,
>
> Index Color
> -------------------
> 0-6 TRANSPARENT
> 7-10 YELLOW
> 11-25 RED
>
> However, the LUT does not provide an alpha channel, except globally
> via Relative Opacity (0070,0403).
>
> Thus, I am stuck. I don't see how we can indicate transparent (non-
> colored) areas in the overlay.
>
> Scott

David:

Thanks for your reply. It provided the clarification I needed.

The blending equation in PS 3.4 N.2.4.3 for each color component, c,
is:

c_out = c_super * A + c_under * (1-A).

This makes it clear to me that BSPS is limited to "blending," as the
name implies. I was thinking that it might have additional
capabilities, i.e., general compositing in the sense of Porter and
Duff.

In a general compositing framework, in addition to the color at each
pixel, there is an alpha value (á). So for two images A & B, we have
color and alpha: c_A, á_A, c_B, á_B. Furthermore, fractions f_A and
f_B denote the fractional amount that each pixel contributes to the
composited pixel. The resultant color and alpha are then:

c = f_A * c_A + f_B * c_B
á = f_A * á_A + f_B * á_B

So in addition to blending, many effects are possible within this
framework.

For example, my goal was to display the underlying grayscale image (A)
unless the CAD probability of malignancy (image B) was above some
threshold, in which case an appropriate opaque color would overwrite
the pixel. In the composite framework, I could be achieved by
assigning transparent black (R,G,B,á=0) for low probability pixels in
image B, and opaque colors for those pixels I wanted to show up. Then
by setting the fractions f_A = (1-á_B), f_B=1, the resulting color and
alpha values gives the desired effect:

c = (1-á_B) * c_A + c_B
á = (1-á_B) * á_A + á_B

Do you know of some other way to achieve this?

Are you looking for someone to write up "CSPS: Compositing Softcopy
Presentation State"? ha, ha.

Regards,
Scott