Dots Detection

Suppose you're looking at a nucleus holding a few bright FISH signals. Each signal is a small, roughly round bump of intensity. Rather than thresholding intensity, the Dots (FISH/CISH) engine tests at every pixel whether the neighborhood matches a dot.

1. Pattern generation. The engine generates a series of dot patterns from your Dot Size Range and Windows Size Range. The reference figures show patterns with the window size varied at a fixed dot size of 7, and with the dot size varied in a window of 25.
2. Matching. Each pattern searches the source image and produces a matching score for every pixel.
3. Selection. The scores from all patterns are evaluated against Sensitivity. A lower Sensitivity returns more dots.

A Background Value can be subtracted from the source image to attenuate high background. If you connect an Events Mask, such as detected nuclei, the search runs only inside those events. The output is a set of events in a coded image, where every pixel of a dot carries that dot's unique ID, so touching dots keep their own shapes.

You add the engine with Add > Detection > Dots (FISH/CISH), in either the Pre-Processing or the Detection section of a layer. It detects; it doesn't measure. Counting and intensity come from Dot MeasurementsLoading....

Why every dot looks about the same size. In light microscopy generally, an emitter smaller than the resolution limitLoading..., about 200 to 250 nm for visible light at high numerical aperture, is imaged as a blur set by the optics: the point spread functionLoading.... A single probe spot and a somewhat larger particle can look alike. A dot's apparent size tells you about the microscope and the pixel size, not about the target.

Why pattern matching suits that. If real dots share nearly one shape, a natural detector compares the image with that shape. In image processing generally this is template matching, a cousin of the Laplacian-of-Gaussian blob detectors used in many spot counters, such as those built for smFISHLoading.... StrataQuest's reference describes its own version, generated patterns over size ranges scored per pixel, and neither a Laplacian-of-Gaussian step nor a link to a measured point spread function. The two ranges are how you tell the engine what your dots look like.

Why a dot must stand above the noise. A rule of thumb from imaging science, the Rose criterion, says a feature is reliably seen when its signal is about five times the noise. A dim probe can fail that test whatever detector you use. Lowering Sensitivity then returns more detections, and in general more of them are noise. Signal-to-noise ratioLoading... is mostly decided at acquisition.

When dots merge. Two signals closer than the resolution limit appear as one brighter spot, and the reference describes no step that splits them. Dot Measurements does report Dots Sum Intensity and a Dots Intensities List per nucleus. If you know the brightness of a single dot, you can estimate how many dots a bright spot holds; that is your calculation, not the engine's, and photon noise makes it approximate.

The reference lists no default values for these settings. Save Profile stores all detection settings as a profile, Load Profile applies one in another project, and several predefined profiles come with the software. Background Value subtracts one number; the Background RemovalLoading... engine instead subtracts a modeled background, which suits backgrounds that vary across the image.

Source: StrataQuest 8.1 Engines Reference, ch. 17 (pp. 417–421), ch. 18 (pp. 422–425), ch. 9 (p. 379), ch. 5 (p. 15).

Here is how a HER2 FISH count could be built. The engine names are StrataQuest's; the scoring rule at the end is the published ASCO/CAP convention, not part of the software.

1. Detect nuclei on the DAPI channel with NucleiLoading... detection.
2. Run Dots (FISH/CISH) on the HER2 probe channel, with a Dot Size Range and Windows Size Range suited to the probe or a predefined profile loaded. Use the nuclei as the Events Mask.
3. Run Dots (FISH/CISH) again on the CEP17 centromere probe channel.
4. If the dots and nuclei were detected in different layers, bring the dots into the nuclei layer with the Coded ImageLoading... import engine.
5. Add Dot Measurements once per probe, with the nuclei as Coded, the probe channel as Shades and that probe's dots as Dots Coded, and select Dots Count.
6. In the Statistics Report, sum each Dots Count over the nuclei and divide the HER2 sum by the CEP17 sum.

The published ASCO/CAP convention uses a HER2/CEP17 ratio of 2.0 as the main dividing line for amplification, with further rules based on HER2 copy number. To see which individual nuclei exceed a ratio, compute it per nucleus with Derived MeasurementsLoading... and place a cutoffLoading... on a histogram of it. Software applies the same settings to every nucleus, which manual scoring can't promise.