Laser speckle contrast imaging turns a nuisance into a measurement. Illuminate rough tissue with coherent light and you get a grainy interference pattern that obscures the image. Let that pattern shift during the camera exposure and the graininess blurs out. How much it blurs tells you how fast things are moving underneath, which is how LSCI produces a full-field map of blood flow in a single frame without scanning and without contrast agents.

The method is cheap in hardware and unforgiving in detail. Most of the detail sits in the light source and the sampling, so that is where this article goes.

Why rough surfaces make speckle

A surface that is rough on the scale of the wavelength scatters coherent light into many wavelets with random relative phases. Where they meet at the detector they interfere, constructively in some places and destructively in others. The result is the familiar grainy pattern, and it is fully deterministic: as long as nothing moves, the pattern stands still.

The same physics that makes speckle a problem in holography and in triangulation sensing makes it the signal here. Move the scatterers, in tissue mostly red blood cells, and the pattern reorganises. The characteristic time over which it becomes unrecognisable is the decorrelation time τ_c, and it falls as the scatterers speed up.

Contrast, exposure and decorrelation

The camera integrates over an exposure time T. If T is much shorter than τ_c, the pattern barely moves and you record it at full contrast. If T is much longer, you average many independent patterns and the graininess washes out. Speckle contrast is defined as the ratio of standard deviation to mean intensity within a small window:

T / τc	Speckle contrast K (at β = 1)	What it means
1	0.75	exposure comparable to decorrelation, little averaging
5	0.42	a useful working point
10	0.31	a useful working point
50	0.14	contrast approaching the noise floor
100	0.10	little dynamic range left for differences in flow

Above roughly ten the relation simplifies to K ≈ √(τ_c/T), which is the form most papers quote. Reported decorrelation times in tissue run from about 0.02 to 0.2 ms in vessels and 0.2 to 0.5 ms in the capillary bed, which puts practical exposures in the range of 1 to 10 ms. The usual guidance is to sit at a few decorrelation times rather than at a hundred, because that is where a change in flow produces the largest change in contrast.

Note what this means for a single fixed exposure: a setting that resolves capillary flow well is already saturated in a large vessel, and one tuned to the vessel is insensitive in the capillary bed. Multi-exposure variants of the method exist precisely because of this.

How much coherence you need

Here LSCI differs sharply from interferometry. There the path difference is set by your geometry and can run to metres. Here the relevant path differences are those the light accumulates inside the tissue before it comes back out, and they are far shorter. But they are not negligible, and the threshold is higher than it was long assumed to be.

The older assumption was that anything above roughly 100 µm of coherence length would do, which at 785 nm corresponds to about 2 nm of linewidth. Measurements have moved that figure: a linewidth below about 0.5 nm, so a coherence length of roughly 400 µm, is the published recommendation.

Source at 785 nm	Linewidth	Coherence length	Suitable for LSCI
free-running Fabry-Perot diode	~2 nm	~98 µm	below the recommended threshold
at the published threshold	0.5 nm	392 µm	the minimum to aim for
VHG-stabilised diode	~0.1 nm	~2 mm	comfortable
wavelength-stabilised module	50 MHz	1.9 m	more than the method needs

What does the difference buy you in practice? Published side-by-side measurements of a stabilised against an unstabilised 785 nm diode report global speckle contrast of 0.83 against 0.59. That is a factor of 1.41, so about 40 % more contrast from the same optical arrangement, purely from the source. Since the flow signal rides on that contrast, it propagates straight into the dynamic range of the measurement.

A second effect is worth separating from the first. Delivering light through a multimode fibre scrambles the spatial coherence and lowers the contrast independently of the linewidth. If your illumination arm uses multimode fibre, source coherence matters more rather than less, because you have already spent part of your contrast budget elsewhere.

Why 785 nm dominates

Most LSCI systems work between 780 and 800 nm, and the reason is tissue rather than optics. In that window haemoglobin absorption has fallen well below its visible-range values while water absorption has not yet risen, so light penetrates further and more of it comes back out. Silicon sensors still work there, which keeps the camera inexpensive, and 785 nm modules are widely available because Raman instrumentation uses the same wavelength.

Wavelength	Argument for	Argument against
532 nm	strong absorption by haemoglobin gives high vessel contrast in the image	shallow penetration, and the same absorption limits how much returns
633 nm	visible, useful for alignment and for exposed-tissue work	more absorption than the near infrared
785 nm	the biological window: good penetration, silicon detectors still sensitive	invisible, so aiming needs a camera or a pilot beam
Above 900 nm	deeper penetration still	silicon sensitivity falls away, water absorption rises

Shorter wavelengths are not simply worse. Where the question is which vessels sit at the surface rather than how fast blood moves two millimetres down, green illumination gives a more contrasted anatomical picture. It is a different measurement, and some systems use both. Where the system also does fluorescence imaging, the choice of LSCI wavelength has to stay clear of the emission bands you want to detect.

Speckle size and the camera

This is the part most often got wrong, and it is invisible in the result: the contrast simply comes out too low and the system looks less sensitive than it is.

Speckle size at the sensor is set by the imaging optics, not by the tissue:

At 785 nm and unity magnification, f/8 gives speckles of about 15 µm, f/4 about 8 µm. If the pixel is larger than the speckle, each pixel averages several independent speckles and the measured contrast falls below the true value. The working rule is to keep speckles at two pixels or more; with 5 µm pixels at unity magnification that means stopping down to about f/5 or beyond.

Stopping down costs light, so there is a real trade against exposure time and source power. It is better made deliberately than discovered afterwards.

What to specify

Parameter	What decides it	Guidance
Linewidth	contrast ceiling β	below 0.5 nm at 785 nm; a wavelength-stabilised module puts the figure out of reach of the problem
Wavelength stability	drift changes nothing in LSCI directly, but mode hops change the coherence	mode-hop-free over the working temperature range matters more than absolute accuracy
Power stability	contrast is a ratio, but drift across a frame sequence shows up as a flow trend	specify RMS over the duration of a measurement, with the bandwidth it was measured in
Output power	field of view, exposure time, f-number, tissue return	tens of mW is common for a few square centimetres; stopping down for sampling raises it
Polarisation	polarised illumination with a crossed analyser suppresses the surface reflection	a defined, stable polarisation state
Beam homogeneity	contrast is calculated locally, but uneven illumination changes the noise across the field	a diffuser or an engineered diffuser, not a raw Gaussian spot

One caution on the camera side. Speckle contrast is a statistic computed over a small window, typically 5×5 or 7×7 pixels, and it is biased by read noise and by quantisation. Working at around 30 % of the dynamic range rather than near saturation is the usual recommendation, which leaves headroom for the bright speckles without pushing the dark ones into the noise.

Frequently asked questions

What does laser speckle contrast imaging actually measure?

It measures how quickly the speckle pattern decorrelates, which is related to the speed of the scatterers. Converting that to an absolute flow rate requires assumptions about the scattering regime, so most systems report a relative perfusion index rather than millilitres per minute.

How much coherence length does LSCI need?

Roughly 400 µm is the published recommendation, which at 785 nm corresponds to a linewidth below about 0.5 nm. That is far less than interferometry needs, but more than a free-running Fabry-Perot diode delivers.

Can I use an LED?

Not for this method. An LED has a coherence length of a few micrometres, so there is no speckle pattern to decorrelate. LEDs are used in LSCI systems for white-light reference images alongside the laser, not instead of it.

Why is my contrast lower than published values?

Three usual causes, in order of how often they turn out to be the one: speckles smaller than two camera pixels, a source with too little coherence, and an exposure time far longer than the decorrelation time. All three reduce contrast and none of them is visible in the raw image.

What exposure time should I use?

Between 1 and 10 ms covers most tissue work, with shorter exposures for large vessels and longer ones for the capillary bed. Picking it from the decorrelation time of the vessels you care about beats picking it from a paper on a different preparation.

Does polarisation matter?

It helps. Illuminating with polarised light and placing a crossed analyser before the camera suppresses the specular reflection from the surface, which carries no flow information and only reduces contrast.