Line scan slide scanning: match line rate to stage speed

Whole-slide digitization is a line scan problem at heart. The slide sweeps under a trilinear camera one bright line at a time, and the stitched image is only as good as the match between stage speed and line rate. Get that match wrong and pixels stop being square — features shrink along the travel axis, colors fringe at every edge, and every measurement taken off the slide drifts. This page gives you the rule that sets stage speed, the arithmetic from 1× macro passes down to 40× detail, and the color-alignment and light budgets that follow from it. The worked numbers use one common whole-slide sensor class — 4096 pixels, 7.5 µm pixel size, 66 kHz maximum line rate, three RGB rows — and the same four formulas carry any other sensor you substitute.

How fast can the stage run at a given line rate?

Hold the stage at or below line rate times object pixel pitch: at 1× magnification a 66 kHz camera with 7.5 µm pixels tops out at 495 mm/s, and the same camera at 20× must slow to 24.8 mm/s.

The camera fixes what it samples across the line; the stage supplies the second axis. Square pixels demand that the slide advance exactly one object pitch between consecutive lines, which pins stage speed to line rate:

Run faster than v_max and the image compresses along travel while each line integrates over more than one pitch of motion — smear no algorithm undoes. Run slower and the image stretches. Set the stage to 80-90 % of v_max so ripple and settling never request lines the camera cannot trigger: at 60 kHz the 20× example holds 60,000 × 0.375 µm = 22.5 mm/s. The camera-side settings behind that number are covered in how to set the line rate of a line scan camera.

What sets the object-space pixel pitch on a slide?

Object pixel pitch is sensor pixel size divided by magnification: 7.5 µm pixels at 20× sample the slide every 0.375 µm, and 5 µm pixels at 40× sample every 0.125 µm.

Magnification drives everything on a slide scanner. With the example sensor (4096 pixels, 7.5 µm, 66 kHz):

Two consequences follow. First, a finer-pixel sensor does not buy speed: 5 µm pixels at 20× sample at 0.25 µm and cap the stage at 66,000 × 0.25 µm = 16.5 mm/s, a third slower than the 7.5 µm sensor at the same line rate.

Second, optics set the resolution floor at high magnification. A 0.5 NA objective resolves about 0.61 λ / NA = 0.61 × 0.55 µm / 0.5 ≈ 0.67 µm, which 0.375 µm sampling covers with roughly 1.8 samples per resolvable detail — sound oversampling for morphology. Aperture also fixes the light the camera sees: f-number runs near 1 / (2 × NA), so NA 0.5 behaves like f/1, and depth of focus sits near λ / NA² = 0.55 µm / 0.25 ≈ 2 µm — the reason slide scanners track focus while scanning.

Counting samples per feature works the same way as grain sorting with a line scan camera: pick pitch from the smallest feature you must resolve, then let magnification follow.

How wide is one swath, and how long does a slide take?

One swath spans line pixels times object pitch: 4096 pixels at 0.375 µm covers 1.54 mm, so a 15 mm tissue area needs 10 passes and about 6 s of pure travel at 24.8 mm/s.

The arithmetic for a 15 × 15 mm tissue region at 20×:

• Swath: 4096 × 0.375 µm = 1.536 mm
• Passes: 15 mm / 1.536 mm = 9.8, so 10 with stitching overlap
• Travel per pass: 15 mm / 24.75 mm/s = 0.61 s
• Pure imaging: 10 × 0.61 s ≈ 6.1 s

The stage spends more than that on step-and-settle between passes, focus mapping and reference lines, so expect whole slides in tens of seconds — overhead dominates, not imaging. Pass count is the lever: a swath twice as wide halves the passes, which is the case for long lines and the lowest magnification the diagnosis allows.

One geometry note: an infinity-corrected 20× objective in a 200 mm tube works out to a 10 mm focal length (200 mm / 20) and sits a few millimetres from the cover glass; long-working-distance types trade aperture for tens of millimetres of clearance. The geometry behind swath, sensor and working distance is set out in line scan camera field of view: how to calculate it.

How do you keep red, green and blue aligned?

Trigger every line from the stage encoder and delay the two earlier color channels by the row separation expressed in lines; distance-based triggering keeps that count fixed even when the stage accelerates.

A trilinear sensor lays its red, green and blue rows along the travel axis, so one point on the slide crosses them at three different moments. Recombination is a delay equal to row separation divided by pixel pitch, in lines. With an encoder pulse per unit distance, every line marks equal travel and the delay is a constant line count, correct at any stage speed.

Free-run the camera at a fixed rate instead, and any speed ripple shifts the channels by the ripple fraction of the row offset — a permanent color fringe on edges running along travel. Cameras in this class add sub-pixel interpolation between channels and tilt correction for a line not perfectly square to travel; both refine that delay, neither replaces encoder lock. To keep the rate glued to the stage over a whole scan, see make the line rate follow machine speed at runtime.

How do you prove the pixels are square?

Scan a round target and measure both axes: an along-track diameter 5 % longer than cross-track means the stage runs 5 % slow, and that aspect error lands in every measurement you take.

Three checks, cheapest first:

1. Circle test. Scan a printed circle or a microsphere array. A circle returns as an ellipse when the axes disagree, and the axis ratio is the speed error to correct directly.
2. Known square. Count pixels across a stage-micrometer square in both directions and match the counts to 1 % before trusting dimensions.
3. Color edge. Zoom a red-on-white edge; a fringe wider than one pixel means the trilinear delay is short by that many lines.

The full geometric route is in line scan camera calibration with a moving checkerboard target. Before any of it, settle the encoder: the scale must deliver more than one count per line at your pitch, the arithmetic in encoder ticks per pixel for line scan — at 0.375 µm per line that demands sub-micrometre scale resolution, which precision linear stages carry natively.

How much light does a 15 µs line need?

Exposure can never exceed the line period, which is 15.2 µs at 66 kHz, so a fast slide scanner lives or dies on light: bright sources, TDI summation, or both.

At 24.75 mm/s with full-period exposure the slide moves 24,750 µm/s × 15.15 µs = 0.375 µm during integration — one whole pixel of smear. Hold exposure to half the period and smear halves with it, at the price of light. The budget then closes three ways: brighter illumination, lower speed (halve the speed and the period doubles), or TDI, which sums successive lines on the sensor and multiplies signal by the number of summed lines without adding smear.

Sizing the source is exactly the problem line scan light: how much power a fast line needs works through. Strobed overdrive belongs behind a controller, as lighting controllers for line scan cameras explains, and binning trades resolution for signal where a dimension allows it — pixel binning on a line scan camera.

Which camera, lens and light do you start from?

Start from a 4K color line scan camera with 7 µm-class pixels for the imaging line, a TDI camera where light is short, and microscope objectives at 20× or 40× for magnification; line scan lenses cover the low-magnification macro passes.

The imaging line first:

The MV-CL042-91GC-V2 is a 4K color line scan camera on GigE with 7 µm pixels, an M42 mount and a 29 mm image circle. At 20× its sampling is 7 µm / 20 = 0.35 µm per pixel, so every 60 kHz of line rate holds 60,000 × 0.35 µm = 21 mm/s of stage; a single color pass keeps the stage sweeping one way.

Where light is short:

The LS-LCM-0016 is a 2K GigE line scan camera with 7 µm pixels, TDI and a 58 kHz line rate. At 20× the same 0.35 µm pitch puts that rate at 58,000 × 0.35 µm = 20.3 mm/s, with TDI stacking several lines of signal into each output line — the trade for dim fluorescence.

For wide macro passes over gross specimens and slide trays, add these:

The MV-CL166-91F2C is a 16K line scan camera on XoFLink with 5 μm pixels, RGB with 4TDI and an 82 mm image circle. At 20× it samples at 5 µm / 20 = 0.25 µm and lays down 16,384 × 0.25 µm = 4.1 mm of tissue per pass — roughly a third of the pass count of a 4K line.

The LS-LSL-0215 is a 25 mm, 0.24x line scan lens in M42 with a 30-45 mm image circle. On 7 µm pixels that magnification samples at 7 µm / 0.24 ≈ 29 µm per pixel, so a 4K line spans roughly 119 mm of tray in one sweep — enough to take a whole slide tray row at once.

Match the lens image circle to the sensor line length before anything else — line scan lens selection for 4K, 8K and 16K cameras pairs them by series. For illumination, choose from line scan lights a fixture at least as long as your widest swath, with fan cooling when the duty cycle runs high.

When is line scan the wrong answer for slides?

Choose area scan when the slide sits still, when exposure must run far longer than a line period, or when you need focal stacks; line scan wins on throughput over a moving stage.

Three honest cases. A bench setup digitizing a handful of slides a day stitches area-scan tiles more simply than it builds a precision stage — and an area sensor can borrow line-scan behavior when parts do move, as area scan camera line scan mode for moving-web inspection shows.

Dim fluorescence wanting second-long exposures defeats any line period: TDI narrows the gap, but a cooled area sensor taking its time wins below a certain photon budget. Z-stacks for 3D reconstruction are an area-scan job outright. Above a few dozen slides a day, or wherever color must survive speed, the moving line takes it.

What should you check before ordering parts?

Run the numbers in this order: object pitch, stage speed, swath, line-rate headroom, light, encoder — and if one fails, change magnification or camera before you buy.

• Object pitch = pixel size / magnification. Fix magnification first; it moves every number after it.
• Stage speed = 0.8-0.9 × line rate × object pitch, with the stage holding it to about 1 % ripple.
• Swath = line pixels × object pitch; passes = tissue width / swath. Fewer passes beats a faster stage.
• Exposure ≤ line period, budgeted at half the period to halve smear; TDI where light runs out.
• Encoder: more than one scale count per line at the final pitch.
• Image circle of the lens at least the sensor line length, and the mount matched to the camera.
• Link budget: line rate × pixels × bytes per pixel within the interface — a 4K color line at 66 kHz and 3 bytes per pixel is 4096 × 3 × 66,000 ≈ 0.81 GB/s, past GigE and into Camera Link or 10GigE territory; see Camera Link frame grabber compatibility for line scan.

Work the sequence in the line scan camera finder, start a matched build at the configurator, and bring borderline numbers to us through the contact page.