Phase-Contrast Imaging Techniques

• Phase-contrast imaging is a family of wave-optical methods that generate image contrast via phase shifts rather than absorption, enabling visualization of transparent samples.
• It employs diverse architectures—including propagation-based, grating, and analyzer-based systems—to convert subtle phase variations into measurable intensity changes.
• Quantitative phase retrieval and 3D tomography techniques enhance imaging accuracy, achieving low-dose, high-speed measurements for biomedical and materials applications.

Searching arXiv for recent and foundational papers on phase-contrast imaging to ground the article. Phase-contrast imaging is a family of wave-optical imaging methods in which contrast arises from sample-induced phase shifts, refraction, or wavefront curvature rather than from attenuation alone. It is therefore particularly effective for transparent optical specimens and for weakly attenuating X-ray specimens such as soft tissue and other low-$Z$ materials, where conventional absorption contrast may be small. In the X-ray setting this distinction is commonly expressed through the complex refractive index

$$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$

with $\delta$ governing refraction and phase shift and $\beta$ governing attenuation; in microscopy it appears as the problem of imaging phase objects whose refractive-index or thickness variations are not directly visible in bright field (Paganin et al., 2020, Paganin et al., 2019).

1. Physical basis of phase contrast

The formal basis of phase-contrast imaging is the complex scalar wavefield, whose intensity and phase jointly determine image formation. In the paraxial regime, phase gradients redirect energy flow, and phase curvature changes intensity during propagation. A central relation is the transport-of-intensity equation (TIE),

$$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$

which links longitudinal intensity variation to transverse phase structure. In this formulation, attenuation contrast is associated with the imaginary refractive component, whereas phase contrast is associated with the real component and becomes visible only after propagation or through a phase-sensitive optical system (Paganin et al., 2019, Paganin et al., 2020).

Propagation-based phase contrast makes this conversion explicit. After a specimen imposes phase shifts, free-space propagation generates Fresnel diffraction, often seen as edge-enhancement fringes at material boundaries. This is especially useful when $\beta$ is small and $\delta$-driven effects are stronger than attenuation differences, as in biological soft tissue and other weakly absorbing specimens (Alloo et al., 2021).

A complementary mechanism is the Zernike principle. Here the outgoing wave is decomposed into an undiffracted reference component and diffracted or scattered components. If the reference is given a relative $\pi/2$ phase shift, interference converts otherwise invisible phase variations into intensity variations. In hard X-rays this principle can be implemented without conventional focusing optics by using dynamical diffraction in a two-block crystal system, with a phase shifter placed in the inter-block gap to act only on the undiffracted beam (Haroutunyan, 26 Feb 2026).

2. Principal imaging architectures

Phase-contrast imaging is not a single modality but a set of architectures that differ in how phase is encoded, sampled, and reconstructed.

Modality	Contrast-generation mechanism	Representative properties in cited work
Propagation-based	Free-space propagation converts phase shifts into Fresnel intensity modulations	Single-distance PB-PCXI; spectral propagation-based decomposition
Grating-based differential phase contrast	Talbot or Talbot-Lau interferometry converts refraction into fringe shifts	Phase-stepping, integrating-bucket, electromagnetic phase stepping, photon-counting mammography
Analyzer-, speckle-, or Fourier-plane methods	Rocking-curve sensitivity, speckle displacement, or phase-plate interference	ABPCI, SPINNet SBI, Zernike-type microscopy and hard-X-ray crystal systems

Propagation-based X-ray phase-contrast imaging (PB-PCXI) records projections at a sample-to-detector distance where phase gradients have evolved into measurable intensity variations. In its tomographic form, a projection is recorded at one distance for many rotation angles, a phase-retrieval algorithm is applied to each projection, and standard filtered back-projection reconstructs the volume. The single-distance, single-exposure-per-angle character is particularly important for radiosensitive samples (Alloo et al., 2021).

Grating-based systems measure differential phase contrast by sampling periodic interference fringes. In Talbot or Talbot-Lau geometries, a phase gradient shifts the fringes laterally, while mean transmission and visibility changes provide absorption and dark-field channels. This architecture underlies compact-source phase contrast, differential phase mammography, and several strategies for replacing conventional mechanical phase stepping with continuous or electronic alternatives (Wali et al., 2017, Miao et al., 2013, Fredenberg et al., 2021).

Analyser-Based Phase-Contrast Imaging (ABPCI), also called Diffraction Enhanced Imaging, uses a perfect analyser crystal operated on a rocking curve. Small refraction-induced angular deviations shift the beam on the rocking curve and are converted into measurable intensity changes. In its conventional form ABPCI is only sensitive to phase gradients lying in the diffraction plane of the analyser crystal, which motivates inclined-geometry extensions for two-dimensional sensitivity (Chalmers et al., 2020).

Speckle-based imaging infers phase from the displacement of a reference speckle pattern caused by the sample. In the near field, the sample and reference intensities satisfy

$$I_s(x, y) = T(x,y)\, I_r(x - \delta x, y - \delta y),$$

with $\delta x,\delta y$ proportional to the projected phase gradient. This approach offers high spatial resolution, strong phase sensitivity, low coherence requirements, and experimental flexibility, but conventional correlation-based reconstruction has been computationally costly (Qiao et al., 2022).

Optical phase-contrast microscopy includes both qualitative and quantitative variants. A modified Gerchberg-Saxton algorithm initialized from a single-shot phase-contrast image was shown to recover quantitative phase for binary and continuously varying phase objects, effectively creating a single-shot quantitative phase-contrast microscope (Hai et al., 2020). A distinct route uses light-field microscopy to synthesize defocused planes from a single exposure and solve the TIE without mechanical $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$0-scanning (Davis, 2012).

Phase-contrast ideas also extend outside conventional specimen imaging. A coherent optical phase-contrast system has been used for tomographic imaging of transient acoustic pressure fields in water by shifting and attenuating the zero-order Fourier component so that the recorded intensity is approximately linear in pressure under the weak-interaction assumption (Clement et al., 2014).

3. Quantitative phase retrieval and inverse problems

A persistent distinction in the field is between qualitative phase contrast and quantitative phase recovery. Conventional phase-contrast microscopy increases visibility but does not by itself provide numerical phase values, and classical Zernike phase contrast “inextricably combines phase information with amplitude” (Davis, 2012). More generally, phase retrieval is the inverse problem of inferring attenuation, phase, or refractive properties from intensity-only data (Paganin et al., 2019).

In TIE-based quantitative phase microscopy, the axial intensity derivative is used to recover phase from defocused images. Under the approximation of constant in-focus intensity, the TIE reduces to a Poisson equation for $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$1, which can be solved efficiently in Fourier space. Light-field microscopy was proposed precisely to supply the defocus diversity required by TIE from a single capture, improving temporal resolution for dynamic transparent samples (Davis, 2012).

A different strategy is to use a phase-contrast image as the initialization for iterative phase retrieval. In the modified Gerchberg-Saxton formulation,

$$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$2

with constant $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$3. Because the phase-contrast intensity already encodes phase information, the search space is reduced and the algorithm converges in only a few iterations; the reported experimental cases converged in about 6 iterations (Hai et al., 2020).

In X-ray PB-PCXI, single-distance Paganin-type retrieval uses the ratio

$$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$4

for a homogeneous object. For multi-material samples this global ratio is only approximate. A notable extension treats the use of an incorrect $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$5 not as a failure but as a measurable deformation of interface shape. For adjacent materials $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$6 and $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$7, the interface-specific ratio is

$$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$8

and an incorrectly retrieved interface is modeled by an error-function-plus-Gaussian profile. Fitting line profiles with a Levenberg–Marquardt algorithm allows the correct interface-specific $$n(\mathbf r)=1-\delta(\mathbf r)+i\beta(\mathbf r),$$9, and hence true $\delta$0 and $\delta$1, to be inferred without a priori sample information (Alloo et al., 2021).

Another line of work formulates phaseless coherent diffractive imaging as a sparse inverse problem with structured illumination diversity. There the object transmissivity

$\delta$2

is reconstructed from multiple masked illuminations and multiple detector measurements. The algorithm explicitly separates incoherent and coherent contributions to intensity, using the former to recover strong absorbers and the latter to recover weak or semi-transparent phase objects. It is stated to have computational cost linear in the number of pixels and to guarantee exact recovery if the image is sparse with respect to a given basis (Moscoso et al., 2022).

4. Tomography and three-dimensional reconstruction

Phase-contrast tomography extends projection methods to volumetric recovery of refractive properties. In PB-PCXI, one projection is acquired per tomographic angle at a single sample-to-detector distance, phase retrieval is applied to each projection, and filtered back-projection reconstructs a three-dimensional CT volume. Applied to a breast-tissue sample containing polypropylene, adipose tissue, glandular tissue, and air, this workflow recovered the refraction component $\delta$3 with 0.6 - 2.4\% accuracy compared to theoretical values (Alloo et al., 2021).

ABPCI tomography has historically been limited by one-dimensional phase sensitivity. An inclined-geometry Laue-analyser method addressed this by rotating the detector and object about the optical axis by $\delta$4, allowing the analyser’s sensitivity direction to mix the object’s $\delta$5 and $\delta$6 phase-gradient components. Using 360-degree tomographic data from a multi-material phantom, the real and imaginary parts of the refractive index, $\delta$7 and $\delta$8, were reconstructed (Chalmers et al., 2020).

Speckle-based tomography has also become quantitative. SPINNet replaces explicit correlation-and-integration with a learned end-to-end inverse model that predicts $\delta$9, $\beta$0, and transmission $\beta$1 directly from a reference and a sample speckle image. The method was demonstrated on 3D X-ray phase-contrast tomography by processing 1800 projections of a flour bug and reconstructing both 3D transmission and 3D phase volumes using filtered back projection (Qiao et al., 2022).

Clinical-scale geometry introduces a separate problem: the grating field of view is much smaller than the desired CT field. A phase-sensitive region-of-interest CT architecture addresses this by embedding a small grating-based PCI system within a full-field absorption CT. Full-field absorption is used for segmentation; local phase measurements calibrate material-specific $\beta$2; missing phase data outside the ROI are extrapolated to correct truncation artifacts and recover quantitative phase values. In simulation the reported metrics improved from RMSE $\beta$3, SSIM $\beta$4 for truncated reconstruction to RMSE $\beta$5, SSIM $\beta$6 after estimation; on mouse data the reported values improved from RMSE $\beta$7, SSIM $\beta$8 to RMSE $\beta$9, SSIM $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$0 (Felsner et al., 2018).

In situ brain PCXI-CT emphasizes the same point in a preclinical setting. Propagation-based phase contrast, combined with a generalized two-material phase-retrieval algorithm adapted for CT, was shown to visualize soft tissues of the brain in situ without contrast agents at a dose $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$1400 times lower than would be required by standard absorption contrast CT (Croton et al., 2018).

5. Noise, dose, speed, and task dependence

Practical deployment of phase-contrast imaging is often limited less by contrast physics than by acquisition speed, dose, stability, and noise propagation. In grating interferometry, conventional phase stepping requires discontinuous mechanical motion and multiple exposures per cycle. The integrating-bucket phase modulation technique replaces this with continuous motion plus time-integrated acquisition; the same phase retrieval formula can then be used, while fast measurement and low dose are described as promising (Wali et al., 2017). Electromagnetic phase stepping removes mechanical grating translation altogether by using a magnetic field to shift the X-ray focal spot. In the reported setup the response time was about 200 microseconds, the focal spot shift reached about 380 $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$2m, and a magnetic field of about 2.4–3.1 mT shifted the projection by one moiré period (Miao et al., 2013).

Speckle-based methods historically traded flexibility for heavy computation. SPINNet altered that trade-off by predicting displacement maps and transmission directly. The reported performance includes phase and amplitude recovery in about 100 ms per image pair, 0.16 s inference for a full $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$3 image pair on one NVIDIA A100 GPU, 0.13 s with a slightly larger batch size, and speeds down to the 50 ms range in a lighter configuration. The residual phase error was about 0.15 rad, or $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$4, close to the theoretical setup sensitivity of $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$5 (Qiao et al., 2022).

Noise behavior depends strongly on modality. In mammography, differential phase contrast was not found to be universally superior to absorption contrast; rather, detectability was task dependent. It improved detectability particularly for fine tumor structures, and the optimal incident energy was higher in differential phase contrast than in absorption contrast. Phase contrast did not facilitate spectral material decomposition, although phase information may be used instead of spectral information (Fredenberg et al., 2021).

A broader noise-analysis framework comparing spectral imaging, differential phase-contrast imaging, and spectral differential phase-contrast imaging (SDPC) focused on projected electron density. The reported optimized variances were $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$6 for spectral imaging, $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$7 for DPC imaging, and $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$8 for SDPC imaging. The same study concluded that conventional DPC produces long-range noise correlations because phase integration amplifies low-frequency noise, whereas SDPC avoids these long-range correlations over a large range of clinically relevant pixel sizes (Mechlem et al., 2019).

6. Applications, limitations, and emerging directions

The application range of phase-contrast imaging is unusually broad. At the Matter in Extreme Conditions endstation of LCLS, a dedicated PCI instrument combines magnified phase-contrast imaging with simultaneous X-ray diffraction for studies relevant to High-Energy-Density Science. It can image phenomena with a spatial resolution of a few hundreds of nanometers and with a temporal resolution better than $$-\nabla_\perp\cdot\left[I(x,y,z)\nabla_\perp \phi(x,y,z)\right] = k\frac{\partial I(x,y,z)}{\partial z},$$9, enabling experiments on shock fronts, phase transitions, void collapses, spallation, hydrodynamic instabilities, and warm dense matter (Nagler et al., 2016).

In biomedical X-ray imaging, phase contrast has been coupled to spectral material decomposition so that propagation-based phase retrieval and basis-material separation occur in a single step. For two-component samples composed of pure aluminium and poly(methyl methacrylate), clean material decomposition without residual phase contrast effects was reported, and in a rabbit kitten’s lung the method produced an un-obstructed image of the lung, free of bones (Schaff et al., 2019). Breast-tissue PB-PCXI similarly demonstrated quantitative recovery of refraction and attenuation without a priori knowledge of sample composition (Alloo et al., 2021).

Several limitations are recurrent. Conventional ABPCI is intrinsically one-dimensional until the geometry is modified (Chalmers et al., 2020). Conventional phase-contrast microscopy is typically tied to a single fixed phase modulation scheme, so bright and dark rendering cannot be adaptively switched for different phase patterns (Feng et al., 26 Apr 2025). Quantitative phase microscopy can suffer phase-wrap ambiguity when optical thickness exceeds a wavelength scale, in which case a phase-unwrapping step would be necessary (Hai et al., 2020). In hard-X-ray Zernike-type imaging with a two-block crystal system, the maximum detectable inhomogeneity size is limited by the entrance slit width (Haroutunyan, 26 Feb 2026).

Recent work pushes the field in two distinct directions. One is information-theoretic generalization. Zernike phase-contrast imaging has been analyzed as a measurement strategy that, in the ideal weak-scattering setting, reaches the quantum limit; even with a finite-width incident beam and a finite-cutoff $\beta$0 phase plate, the reported performance remains $\beta$1 of the quantum limit, and the formalism was extended in principle to arbitrary coherent incident beams, including speckle beams (Dwyer et al., 16 Feb 2026). The second direction is adaptive quantum and metasurface-enabled imaging. A metasurface-assisted scheme combining a polarization-entangled light source with a polarization-multiplexed metasurface realized remotely switchable bright-phase-contrast and dark-phase-contrast imaging. Reported image contrasts were 0.28 for classical imaging, 0.36 for quantum imaging without metasurface modulation, 0.75 for quantum bright-phase contrast, and 0.81 for quantum dark-phase contrast; the system was also reported to image low phase-gradient targets with phase difference as low as $\beta$2 and onion epidermal cells under low-throughput light conditions (Feng et al., 26 Apr 2025).

Taken together, these developments show that phase-contrast imaging has evolved from a qualitative visibility-enhancement technique into a quantitatively reconstructive, modality-spanning framework. Depending on the implementation, it can recover $\beta$3 and $\beta$4, projected phase, projected electron density, material thickness, or signed acoustic pressure; it can operate in single-shot, single-distance, multi-energy, or tomographic modes; and it increasingly integrates optics, computation, and, in some cases, quantum state control.