Doppler Holography Imaging

• Doppler holography is an optical imaging technique that combines digital holography and Doppler velocimetry to measure local blood flow dynamics.
• It employs interferometric setups with high-speed detectors and advanced spectral processing to produce detailed, noninvasive flow maps in biological tissues.
• The method enables pixelwise flow quantification and arteriovenous mapping, offering valuable insights into vascular health and disease biomarkers.

Doppler holography is an interferometric, full-field optical imaging technique that enables quantitative, high-resolution measurement of local motion—primarily blood flow—by combining the spatial phase sensitivity of digital holography with the frequency-resolving power of Doppler velocimetry. It operates by interferometrically recording temporal fluctuations induced by moving scatterers onto a pixel array and reconstructing spatially resolved power spectral densities via advanced digital signal processing. This paradigm allows wide-field, noninvasive measurement of flow velocity, pulsatility, and arteriovenous mapping in living tissue, with applications spanning ophthalmology, microvascular research, and dynamical systems.

1. Physical Principles and Interferometric Architecture

Doppler holography relies on coherent detection within a Mach–Zehnder or Michelson interferometric configuration. A stable, single-frequency laser (commonly λ = 785–852 nm for ophthalmoscopic use) is split into an object (illumination) arm and a reference arm. The object arm illuminates the sample (e.g., retina) and collects backscattered (or transmitted) light, which carries Doppler frequency shifts imparted by moving scatterers, notably erythrocytes. The reference arm provides a spatially and temporally coherent local oscillator (LO). The two beams are recombined (on- or off-axis) on a high-speed detector array, typically a CMOS camera, capturing a sequence of interferograms up to rates of 75 kHz or higher.

The recorded intensity at a pixel (x, y) and time t is

$$I(x, y, t) = |E_{\mathrm{obj}}(x, y, t) + E_{\mathrm{ref}}(x, y, t)|^2,$$

with the cross-term $E_{\mathrm{obj}} E^*_{\mathrm{ref}}$ encoding both amplitude and phase fluctuations due to sample motion.

The system can be operated in off-axis or phase-shifting mode. Acousto-optic modulators (AOMs) in each interferometer arm may introduce a frequency offset Δω, enabling selective frequency down-conversion of high-speed optical fluctuations to the detector bandwidth, and enhanced velocity selectivity by tuning the heterodyne detection frequency (Gross et al., 2012).

2. Doppler Signal Mathematics and Spectral Processing

The core Doppler metric is the frequency shift Δf imparted by moving scatterers, which, in backscattering geometry, is given by

$\Delta f = \frac{2(v \cos \theta)}{\lambda},$

where v is the velocity component along the optical axis, θ is the angle between flow and illumination, and λ is the wavelength.

For each pixel, the complex-valued time series $E(x, y, t)$ is extracted (via numerical propagation and spatial filtering), and its autocorrelation $G(\tau) = \langle E(t)E^*(t + \tau)\rangle_t$ is computed. The power spectral density (PSD) is obtained via the Wiener–Khintchine theorem:

$$S(\omega) = \int_{-\infty}^\infty G(\tau) e^{-i \omega \tau} d\tau.$$

Practically, the PSD is estimated over a short-time window (STFT), enabling temporal tracking of flow dynamics. Integrals of the PSD over selected frequency bands yield power Doppler images:

$$P(x, y) = \int_{\omega_1}^{\omega_2} S(x, y; \omega) d\omega,$$

with differentiated bands revealing slow (venous, tissue) and fast (arterial) flows (Puyo et al., 2019, Puyo et al., 2021). Higher order spectral moments (e.g., $M_n = \int \omega^n S(\omega) d\omega$) provide quantitative velocity metrics.

3. Experimental Realization and Signal Processing Pipeline

A modern Doppler holography implementation for retinal imaging typically involves:

• Mach–Zehnder fiber interferometer at λ = 785–852 nm.
• Object illumination at ∼1.5 mW over a 4×4 mm² field.
• Detection using high-speed CMOS camera (e.g., Phantom V2511: 512×512 px, 12-bit, up to 75 kHz, 13 μs exposure).
• Numerical back-propagation (angular-spectrum or Fresnel method) to obtain the complex field at the sample plane.
• Short-time Fourier transform (STFT) windowing (e.g., 512–1024 frames, T_win ≈ 6.8–13 ms) for Doppler spectrum computation.
• Band integration for power Doppler movies and moment analysis.
• Clutter rejection using SVD-based spatio-temporal filtering to remove coherent eye motion and enhance slow-flow detectability (Puyo et al., 2020).

The entire reconstruction and analysis pipeline is GPU/FPGA-accelerated for near-real-time performance.

4. Power-Doppler Imaging, Index Mapping, and Arteriovenous Discrimination

Doppler holography uniquely enables pixelwise, full-field processing of flow metrics:

• Power-Doppler images (zeroth moment) visualize spatial flow distribution.
• Pulsatility and resistivity indices (PI, RI), calculated as

$$\mathrm{PI} = \frac{P_{\text{peak}} - P_{\text{trough}}}{P_{\text{mean}}}, \quad \mathrm{RI} = \frac{P_{\text{peak}} - P_{\text{trough}}}{P_{\text{peak}}},$$

serve as proxies for vascular resistance and compliance.

• Coefficient of variation (CV) maps, computed per pixel as $\sigma(x, y)/\mu(x, y)$ over the cardiac cycle, distinguish arteries (high CV) from veins (low CV) with >90% specificity (Puyo et al., 2019, Puyo et al., 2021, Dubosc et al., 18 Nov 2025).
• Composite bandpass imaging (e.g., low-frequency band in cyan, high-frequency in red) reliably separates choroidal arteries and veins, with veins dominated by slow flow and arteries by rapid flow components.

Recent work demonstrated that time-resolved features extracted from Doppler movies (e.g., pixelwise cardiac correlation, systole-diastole amplitude) can further boost automated artery/vein (A/V) segmentation using standard neural networks, achieving Dice scores >0.8 (Dubosc et al., 18 Nov 2025).

5. Quantitative Flow Measurement and Model-Based Calibration

Absolute flow estimation is enabled by leveraging deterministic models of Doppler broadening under forward scattering assumptions. For primary in-plane retinal arteries, the RMS Doppler broadening Δf relates to local RMS velocity v via:

$v(x, y) = \frac{\lambda \, \Delta f}{\mathrm{NA}},$

where NA is the numerical aperture of the eye–eyepiece combination (Fischer et al., 24 Sep 2024, Auray et al., 14 Apr 2025). Vessel segmentation (Frangi filtering, correlation with global cardiac signal) yields robust delineation of major arteries. The Poiseuille model (laminar cylinder flow) is fitted to cross-sectional velocity profiles for volumetric flow quantification. Reported total retinal arterial flow in healthy subjects is consistent with gold-standard Doppler OCT and laser Doppler velocimetry (Fischer et al., 24 Sep 2024, Auray et al., 14 Apr 2025).

6. Performance Metrics and Limitations

Metric	Typical Value	Reference
Spatial resolution	5–15 μm (retina), 30-40 μm (choroid)	(Puyo et al., 2019, Puyo et al., 2021)
Temporal resolution	7–13 ms per power Doppler frame (raw ∼13 μs)	(Puyo et al., 2019)
Velocity sensitivity	0.5 mm/s–several mm/s (Doppler shifts 0.2–30 kHz)	(Puyo et al., 2019, Fischer et al., 24 Sep 2024)
Artery/vein mapping acc.	>90% (retina, via CV or temporal cues)	(Puyo et al., 2019, Dubosc et al., 18 Nov 2025)
Flow quant accuracy	±10 μL/min, ≤30% vs. reference	(Fischer et al., 24 Sep 2024, Auray et al., 14 Apr 2025)

Among limitations: lack of intrinsic depth discrimination (en-face images), sensitivity to multiple scattering (especially out of the retinal plane), and computational demands for high-frame-rate acquisition. Temporal decorrelation and phase unwrapping constraints limit accessible velocity ranges at the extremes.

7. Applications, Extensions, and Future Directions

Doppler holography is uniquely positioned for functional vascular imaging in ophthalmology:

• Noninvasive mapping of pulsatile blood flow, hemodynamic indices, and disease biomarkers (e.g., chronic glaucoma, diabetic retinopathy, retinal vein occlusion) (Atlan, 13 Mar 2024).
• Arteriovenous discrimination, resistivity mapping, and detection of pathological flow reversal (Puyo et al., 2019, Puyo et al., 2020).
• Choroidal vascular imaging and dynamic assessment without exogenous dyes, at spatial and temporal scales that surpass OCT angiography and indocyanine green angiography (Puyo et al., 2021).
• Anterior-segment applications, eye tracking, and lens/cornea transparency assessment via retro-illumination reconstructions (Puyo et al., 2021).
• Quantitative research on pulse wave propagation, compliance, and local tissue biomechanics, particularly when integrated with parallel full-field swept-source OCT (Puyo et al., 2021).
• Potential for cross-disciplinary extension to vibrometry, microfluidics, and industrial diagnostics.

Ongoing advances focus on ultrahigh-speed detectors, robust motion-compensation pipelines (spatio-temporal SVD, deep learning), real-time GPU-accelerated reconstructions, and the development of open-access Doppler hologram datasets for community-wide benchmarking (Dubosc et al., 18 Nov 2025).