Compute the axial resolution Δz and lateral resolution Δx of an OCT system in real time from center wavelength, bandwidth and objective NA. Compare SD-OCT, SS-OCT and femtosecond sources, and size the imaging depth, sampling pixels and B-scan time you need for retinal, cardiovascular or skin OCT.
Parameters
Source type
SLD for SD-OCT, Swept Source for SS-OCT, femtosecond for ultra-broad bandwidth
Samples per axial resolution element (≥2, typically 2.5-3)
Sample refractive index n
Soft tissue 1.35-1.4; sclera/tooth ~1.5; water 1.33
Results
—
Axial resolution Δz (μm)
—
Lateral resolution Δx (μm)
—
Depth of focus DOF (μm)
—
Max imaging depth (μm)
—
Required sampling pixels
—
A-scan rate (kHz)
—
OCT sample-arm schematic
The sample-arm beam reaches a layered specimen (e.g. retinal layers); the coherence length ℓ_c sets the depth-resolved interference signal. Band thickness shows axial resolution; cone narrowness shows lateral resolution.
Axial resolution Δz vs bandwidth Δλ
Lateral resolution Δx and DOF vs objective NA
Theory & Key Formulas
$$\Delta z = \frac{2\ln 2}{\pi}\frac{\lambda_0^{2}}{\Delta\lambda},\qquad \Delta x = 0.61\,\frac{\lambda}{NA},\qquad DOF = \frac{\pi\,\lambda}{NA^{2}}$$
Δz: axial resolution (FWHM coherence length of a Gaussian-spectrum source). Δx: lateral resolution (Airy radius). DOF: depth of focus (two-way Rayleigh range). In tissue, divide Δz by the refractive index n to obtain Δz/n.
At the ophthalmologist they said "we'll take an OCT". It's not ultrasound, right — it's light? How can light resolve a 5 μm-thick cell layer?
🎓
OCT (Optical Coherence Tomography) was published by Huang et al. in 1991. It is a tomographic technique that exploits the interference of low-coherence light. The optical principle is a Michelson interferometer: backscattered light from the sample and light from a reference mirror are overlapped, and interference fringes appear only when the two path lengths match. The shorter the coherence length of the source, the narrower the "match" window — and that window directly sets the axial resolution. So an 850 nm SLD with 50 nm bandwidth gives about 5 μm, and an ultra-broadband femtosecond laser reaches 1 μm.
🙋
I see — when I raise "Bandwidth Δλ" on the left, the axial resolution Δz really improves. But raising the objective NA improves lateral resolution while the depth of focus collapses. Why?
🎓
It is a fundamental Gaussian-optics trade-off: DOF = πλ/NA² scales inversely with NA². At NA=0.10 you have a 267 μm DOF, which easily covers the 200-300 μm retinal thickness in one shot. At NA=0.4 only 17 μm of DOF is left, so just one tissue layer is sharp at a time. That is exactly why retinal OCT deliberately uses low NA — engineers accept Δx of 15-20 μm in order to keep all retinal layers in focus.
🙋
So what do people do when they need deep AND high lateral resolution, like skin or dental OCT?
🎓
Three approaches. First, switch to SS-OCT at 1050 nm or 1310 nm where scattering is lower and light penetrates further. Second, use a tunable (liquid) lens to refocus at successive depths and stitch the in-focus slices together — that is Gabor Domain Fusion. Third, use a Bessel beam or axicon, which keeps a near-diffraction-limited spot over a much longer distance. Modern dental OCT and skin OCT (VivoSight) combine these to reach 5-10 μm laterally over 1-2 mm of depth.
🙋
The A-scan rates are different between SD-OCT and SS-OCT. Why? Is faster always better?
🎓
The difference comes from the detection method. SD-OCT uses a broadband source plus spectrometer plus line-scan camera, so camera readout caps it at 70-100 kHz. SS-OCT uses a MEMS swept laser at 200-400 kHz, and the latest VCSEL pushes beyond 1 MHz. Faster is great because it reduces motion artefacts from heartbeat and eye saccades and can grab a 3D volume in one second, but it also reduces the photons collected per A-scan and lowers sensitivity. So fast SS-OCT wins for OCT angiography of retinal blood flow, while slower SD-OCT remains the choice when high sensitivity matters more than speed.
🙋
"Required sampling pixels" comes out around 1500 — is that the camera's pixel count?
🎓
Yes — for SD-OCT it is the number of pixels on the spectrometer line camera, for SS-OCT it is the ADC samples per sweep. Nyquist requires at least 2 samples per axial resolution element; in practice we use 2.5-3 to suppress FFT sidelobes. For a 3 mm depth with 4.7 μm axial resolution and a Nyquist factor of 2.5 you need 1588 samples — so a 2048-pixel line camera fits the bill, and that drives the spectrometer specification.
Frequently Asked Questions
OCT records the autocorrelation of a low-coherence light source as an A-scan. For a Gaussian source spectrum, the full-width at half maximum of the coherence function is the axial resolution Δz=(2ln2/π)·λ₀²/Δλ. In tissue, this is divided by the refractive index n: a wider bandwidth Δλ damps the interference faster, giving finer axial resolution. With center wavelength 850 nm, bandwidth 50 nm and n=1.35 you get about 4.7 μm; with 100 nm bandwidth, about 2.4 μm.
SD-OCT (Spectral Domain) combines a broadband SLD with a spectrometer, typically running at 70-100 kHz A-scan rate around 800 nm with 3-6 μm axial resolution. It is the workhorse of retinal OCT (Heidelberg Spectralis, Zeiss Cirrus). SS-OCT (Swept Source) uses a MEMS swept laser at 1050-1310 nm with 100-400 kHz rates and shallow sensitivity roll-off, making it ideal for deep imaging (choroid, cardiovascular, deep skin). Femtosecond lasers provide ultra-broad bandwidth for 1-2 μm ultra-high resolution at higher cost.
Increasing the objective NA improves lateral resolution Δx=0.61λ/NA, but depth of focus DOF=πλ/NA² drops with the square of NA. At NA=0.10 you get Δx=5.2 μm with DOF=267 μm (the entire retina is in focus); at NA=0.4 you get Δx=1.3 μm but only 17 μm of DOF, so only one tissue layer is sharp in a thick sample. Retinal OCT therefore stays at low NA and accepts Δx<20 μm, while high-resolution skin OCT uses tunable lenses or Bessel beams to recover both.
OCT sensitivity scales with the photon flux returning to the detector, quantum efficiency and integration time (=1/A-scan rate). Commercial SD-OCT systems reach about 90-100 dB and SS-OCT 100-110 dB. Wider bandwidth distributes the same photons over more spectral channels, lowering per-channel SNR (this tool benchmarks against 50 nm bandwidth). Sensitivity also rolls off with depth (6-10 dB/mm for SD-OCT). SS-OCT has a much longer coherence length so its roll-off stays nearly flat at depth.
Real-World Applications
Retinal OCT (ophthalmology): Clinical systems such as Heidelberg Spectralis, Zeiss Cirrus and Topcon Triton lead this field. Operating at 820-870 nm SLD with 3-7 μm axial and 15-20 μm lateral resolution, they non-invasively visualize the ten retinal layers (nerve fiber, ganglion cell, outer plexiform, photoreceptor and so on). They are essential for early diagnosis of diabetic retinopathy, age-related macular degeneration (AMD) and glaucoma, and over one hundred million scans are performed worldwide every year.
Cardiovascular OCT (IVOCT): Abbott's OPTIS (formerly LightLab) is the flagship. A 1310 nm SS-OCT engine at ~100 kHz A-scan rate delivers about 25 μm lateral and 15 μm axial resolution. A catheter is inserted into the coronary artery to visualize stent apposition, thrombus and thin-cap fibroatheroma (TCFA). At an order of magnitude finer than the 100 μm of intravascular ultrasound (IVUS), it is now central to assessing culprit lesions in acute coronary syndromes.
Dermatology & dentistry: VivoSight (Michelson Diagnostics) and Heidelberg DermaSight use 1310 nm SS-OCT to image the epidermis and dermis to 1-2 mm depth. Clinical uses include margin assessment of melanoma and non-melanoma skin cancer and longitudinal monitoring of scar therapy. In dentistry, OCT is gaining adoption for detecting incipient enamel caries (white-spot lesions) and evaluating adhesive-to-tooth interfaces.
Endoscopic OCT and industrial uses: The leading clinical application is dysplasia screening in Barrett's esophagus (NinePoint Medical NvisionVLE) and assessment of lateral margins of gastrointestinal lesions. In industry, "metrology OCT" with μm-scale resolution over mm depth is spreading rapidly: non-contact multilayer thickness measurement of glass, plastic film and coatings; defect inspection of lithium-ion battery separators; and internal-void evaluation of 3D-printed parts.
Common Misconceptions and Pitfalls
The most common mistake is to treat the axial resolution Δz as the free-space value inside tissue. The coherence length of an OCT source does not shrink by a factor of n in a medium; rather, optical depth (geometrical × n) is what is measured, so the effective axial resolution becomes Δz/n. With n=1.35 in retina, an air-equivalent 6.4 μm becomes 4.7 μm in tissue. At the same time, the depth scale is also compressed by n, so leaving the Y-axis label as "optical depth" makes structures look thicker than their true anatomical thickness. Always enable the tissue refractive-index correction provided by the OCT software.
Next, "more sampling means better resolution" — false. Raising the Nyquist factor from 3 to 8 does not improve Δz itself, because Δz is fixed by the coherence length of the source. More samples only suppress FFT sidelobes and artefacts; the true resolution is locked to the physical limit of the optical system. Under-sampling below Nyquist on the other hand produces mirror artefacts (the zero-delay symmetric image) and aliasing, so always keep at least a factor of 2 and ideally 2.5-3.
Finally, watch out for depth specifications that ignore sensitivity roll-off. SD-OCT systems quote a maximum imaging depth of 2-3 mm, but in reality sensitivity drops by 6-10 dB at 1 mm depth and 15-20 dB at 3 mm. That is fine for retinal imaging where the relevant 0-300 μm region is near zero-delay, but applications such as cardiovascular OCT that require uniform sensitivity over 5-10 mm need SS-OCT. The "imaging depth" in a datasheet is the FFT-window upper bound, not a usable sensitivity range — always look at the measured roll-off curve from the vendor.
How to Use
Enter center wavelength (lambdaNum, typical 830–1310 nm for retinal or dermatological OCT)
Retinal OCT system: center wavelength 840 nm, bandwidth 100 nm, NA 0.12, imaging depth 2.0 mm. Axial resolution Δz = 2λ₀²/Δλ ≈ 2(840²)/100 ≈ 14.1 μm. Lateral resolution Δx = 0.61λ₀/NA ≈ 0.61(840)/0.12 ≈ 4.27 μm. Depth of focus DOF = λ₀/(2NA²) ≈ 290 μm. At 2 mm depth with 2048 A-scan pixels and 50 kHz A-scan rate, imaging 512 A-lines per frame yields ~98 Hz B-scan rate, suitable for real-time retinal visualization.
Practical Notes
Increasing bandwidth reduces Δz nonlinearly; doubling bandwidth from 50 to 100 nm halves axial resolution, improving penetration trade-off for choroidal imaging
Higher NA improves lateral resolution but reduces DOF; dermatological systems use NA ~0.2 (DOF ~200 μm) to match skin layer thickness
A-scan rate must exceed 2× (depth/c) × pixel count to avoid aliasing; swept-source OCT at 100 kHz enables 3D volumetric scans (512×512×496 voxels) in <2.5 seconds
For spectral-domain OCT, required sampling pixels scales as λ₀²/Δλ; 1310 nm systems need ~1.7× more pixels than 840 nm for equal resolution