Dynamic Full-Field OCM

• Dynamic Full-Field Optical Coherence Microscopy (d-FF-OCM) is a label-free imaging method that analyzes temporal fluctuations to reveal intracellular activity, enhancing tissue characterization.
• It extends traditional full-field OCT/OCM by using dynamic contrast from interferometric signal variations to distinguish live cellular processes from static structures.
• Variants such as swept-source, rolling-phase, and high-NA deep-tissue implementations demonstrate versatility by addressing motion sensitivity, signal artefacts, and penetration challenges.

Dynamic Full-Field Optical Coherence Microscopy (d-FF-OCM) is a label-free, coherence-gated, full-field interferometric microscopy modality in which repeated en face images or volumes are temporally analyzed to reveal intracellular or intratissue activity that is weak or invisible in static structural FF-OCT/FF-OCM images. In the literature it is commonly described as dynamic full-field optical coherence tomography (D-FFOCT), while high-numerical-aperture, microscopy-scale implementations are also naturally understood as full-field OCM; in practical use, the distinction often depends on application scale and numerical aperture rather than on a different contrast mechanism (Apelian et al., 2016). Recent work has extended the modality from time-domain full-field systems to full-field swept-source OCM, interface self-referenced and rolling-phase variants, and high-NA deep-tissue implementations, while preserving the basic principle that dynamic contrast is computed from temporal fluctuations of coherence-gated interferometric signals (Tateno et al., 13 Nov 2025, Monfort et al., 2023, Tarvydas et al., 5 Aug 2025).

1. Terminology and modality definition

Classical full-field OCT is a white-light interference microscope that produces en face tomographic images by arithmetic combination of interferometric images acquired with an area camera and by illuminating the whole field of view with low-coherence light. In standard FFOCT, the interferometric amplitude is usually recovered by 2-phase or 4-phase stepping of the reference arm. Dynamic full-field OCT extends that architecture by leaving the reference unmodulated during the dynamic acquisition and using the time dependence of the interferometric signal itself to reveal intracellular activity. In that sense, d-FF-OCM is not a separate hardware family so much as a temporal extension of full-field coherence-gated microscopy (Apelian et al., 2016).

The same logic carries over to newer full-field swept-source implementations. In spatially coherent full-field OCM (SC-FFOCM), the structural image is the refocused complex or intensity OCT volume reconstructed from wavelength-swept interferometric data, whereas the dynamic image is a second-order functional map computed from temporal fluctuations across repeated volumes. This structural-versus-dynamic separation is central: morphology is reported by the coherence-gated OCT/OCM channel, while activity-sensitive contrast is reported by temporal fluctuation metrics such as LIV or Late-ODS (Tateno et al., 13 Nov 2025).

2. Physical basis of dynamic contrast

A foundational signal model for D-FFOCT writes the camera intensity as

$$I(x,y,z,t) = I_{0}(x,y,z) + a(x,y,z,t)\cos\!\big(\phi(x,y,z,t)\big),$$

where $I_0$ is the stationary background term, $a$ is the amplitude of the interfering sample field, and $\phi$ is its phase relative to the reference. Dynamic contrast arises because endogenous moving scatterers modulate both $a$ and $\phi$. In the 625 nm system analyzed by Apelian and colleagues, the minimal axial displacement needed to move from a cosine minimum to maximum was

$\Delta z = \frac{\lambda}{4n},$

reported as $111$ nm for $n=1.4$, whereas a transverse displacement of about $750$ nm was required to significantly alter the signal. This anisotropy explains why nanometric intracellular motion can generate strong temporal interferometric modulation (Apelian et al., 2016).

The standard dynamic map in the foundational D-FFOCT paper is the temporal standard deviation

$I_0$0

applied pixelwise to the raw temporal stack. Stationary structures contribute to the mean but not to the fluctuation, so highly reflective yet static components such as collagen, myelin, or capsule fibers are suppressed, while weakly scattering but active cellular regions become visible. The same work reported that fixed tissues showed no D-FFOCT signal, untreated fresh tissue signal decayed over hours, rotenone did not significantly reduce liver signal, and 2-deoxy-D-glucose reduced global liver contrast from $I_0$1 to $I_0$2 within $I_0$3 minutes, consistent with a glycolysis-dependent local activity readout. At the same time, the contrast is not assigned to a single universal biophysical source: in a retinal pigment epithelium stress model, mitochondria were identified as the main contributor to the D-FFOCT contrast, while pigment granules, microvilli, filopodia, lysosomes, and other moving scatterers also contributed characteristic signatures (Apelian et al., 2016, Groux et al., 2021).

3. Instrument architectures and acquisition regimes

The canonical time-domain architecture is a full-field Linnik-type interferometric microscope using broadband illumination and camera-based detection. Foundational D-FFOCT systems employed $I_0$4 water immersion objectives with $I_0$5 or $I_0$6 water immersion objectives with $I_0$7, $I_0$8 axial resolution, tissue-dependent penetration depths of $I_0$9–$a$0, and practical dynamic recordings of $a$1 to $a$2 s. A later module adapted the modality to a commercial Olympus IX83 microscope with a stage-top incubator, $a$3 nm and $a$4 nm LEDs, a $a$5 field of view per tile, theoretical lateral spatial resolution $a$6 nm, and longitudinal imaging of retinal organoids over $a$7 days and over more than $a$8 weeks (Apelian et al., 2016, Monfort et al., 2023).

Several variants were developed to address specific failure modes. Interface self-referenced D-FFOCT (iSR D-FFOCT) blocks the conventional reference arm and uses the top surface of the coverslip as a defocused reference reflector with about $a$9 reflectivity in water. With $\phi$0 objectives and $\phi$1 frames/s acquisition, it measured a transverse PSF of $\phi$2 nm and an axial PSF of $\phi$3 nm at $\phi$4 nm, removed fringe artefacts at $\phi$5, and improved mean-frequency stability by $\phi$6 dB relative to conventional D-FFOCT. Rolling-phase DFFOCT instead continuously scans the reference phase over a range larger than $\phi$7 during acquisition; with $\phi$8 frames at $\phi$9 Hz and a $a$0 or $a$1 ramp, it uses the mean absolute frame-to-frame intensity difference $a$2 as dynamic brightness and the Fourier magnitude at the imposed carrier frequency $a$3 Hz to recover a static image from the same dataset (Monfort et al., 2023, Monfort et al., 14 Jan 2025).

A distinct architectural branch is full-field swept-source OCM. In SC-FFOCM, the sample is illuminated by a plane wave from a wavelength-swept source and a 2D camera records one full-field interferogram per spectral sample. A representative system used a $a$4 CMOS camera, $a$5 interferometric images at $a$6 frames/s per volume, and $a$7 repeated volumes with $a$8 ms repetition and $a$9 s total acquisition time. Its in-focus lateral resolution was $\phi$0, axial resolution $\phi$1 in air, and native depth of focus $\phi$2, with computational refocusing used to extend cellular-resolution imaging through the full thickness of an MCF-7 spheroid (Tateno et al., 13 Nov 2025).

High-NA deep-tissue d-FF-OCM pushed the modality into a different operating regime. A system built around $\phi$3 oil-immersion objectives with $\phi$4, a laser-pumped incoherent white-light source, a $\phi$5 camera at $\phi$6 fps, and real-time reference-arm adjustment reported $\phi$7 nm lateral resolution, $\phi$8 nm axial resolution, a $\phi$9 field of view, and dynamic imaging depths up to approximately $\Delta z = \frac{\lambda}{4n},$0 in highly scattering fresh ex vivo mouse liver and small intestine (Tarvydas et al., 5 Aug 2025).

4. Reconstruction and dynamic observables

In time-domain D-FFOCT, temporal standard deviation and temporal spectral analysis are the dominant reconstruction strategies. In retinal organoid imaging and in commercial-microscope D-FFOCT, each pixel’s temporal trace is processed by Welch PSD estimation, L1 normalization, computation of the mean frequency and spectral bandwidth, and estimation of a running temporal standard deviation over a window of $\Delta z = \frac{\lambda}{4n},$1 samples. These quantities are then mapped into HSV space: hue encodes mean fluctuation frequency, saturation encodes inverse frequency bandwidth, and value encodes fluctuation amplitude, with blue corresponding to slow dynamics and red to fast dynamics (Scholler et al., 2019, Monfort et al., 2023).

Swept-source full-field OCM adds OCT-specific reconstruction before temporal analysis. Raw spectral interference data are rescaled linearly in wavenumber $\Delta z = \frac{\lambda}{4n},$2, Fourier transformed along $\Delta z = \frac{\lambda}{4n},$3, and filtered laterally to suppress stray reflections. Computational refocusing then multiplies the en face spatial-frequency spectrum by the phase-only quadratic filter

$\Delta z = \frac{\lambda}{4n},$4

followed by inverse lateral Fourier transformation:

$\Delta z = \frac{\lambda}{4n},$5

On the refocused 4D dataset, the structural image is separated from the dynamic channel. LIV is defined as the temporal variance of the dB-scale OCT intensity signal, while Late-ODS is the slope of the autocorrelation decay curve within the delay range $\Delta z = \frac{\lambda}{4n},$6, after kernel averaging over $\Delta z = \frac{\lambda}{4n},$7 pixels in axial and lateral directions; in that implementation, Late-ODS was sensitive to speeds around $\Delta z = \frac{\lambda}{4n},$8 (Tateno et al., 13 Nov 2025).

The heavy data burden of full-field dynamic imaging has motivated computational acceleration. In FF-SS-OCM, a neural-network method was trained to predict a high-definition LIV image from only four OCT volumes instead of the conventional $\Delta z = \frac{\lambda}{4n},$9, reducing data size from $111$0 GB to $111$1 GB, transfer time from $111$2 min to $111$3 s, and processing time from $111$4 h to $111$5 min. Because the network is trained against the conventional $111$6-volume LIV image, its output is a learned surrogate of that estimator rather than a physically complete replacement of the underlying temporal process (Komeda et al., 15 Jan 2026).

5. Biological and biomedical applications

The early D-FFOCT literature established the method on fresh ex vivo tissues. In brain cortex, static FFOCT is dominated by bright myelinated axons, whereas D-FFOCT suppresses those stationary structures and reveals cell bodies. In liver, D-FFOCT reveals hepatocyte boundaries and intracellular heterogeneity that are difficult to identify in static images. In kidney, 3D D-FFOCT makes tubules more evident. In mouse intestine, D-FFOCT suppresses stationary collagen and leaves mainly cells in the cancerous zone, including a population interpreted as highly dynamic immune cells (Apelian et al., 2016).

Retinal models became a major application domain. In retinal organoids, D-FFOCT provided 3D live imaging of developing structures, including rosettes and photoreceptor-related regions, and showed that adjacent cell populations such as RPE-like cells and slower progenitor cells could be distinguished by dynamic signature alone. Multimodal validation demonstrated that fluorescently labeled dead cells corresponded to very weak dynamic signals, while CRX-mCherry-positive photoreceptor regions coincided with a distinct endogenous dynamic phenotype (Scholler et al., 2019). In a retinal pigment epithelium scratch-assay stress model, D-FFOCT quantified wound closure and revealed phenotype changes during repair; primary porcine RPE showed small-scratch closure around $111$7, hiPSC-derived human RPE around $111$8, and immunohistochemistry indicated that mitochondria were the main contributor to the D-FFOCT contrast (Groux et al., 2021).

Three-dimensional tumor models motivated volumetric dynamic full-field OCM. In MCF-7 spheroids, SC-FFOCM with computational refocusing showed a low-LIV/low-Late-ODS core and higher dynamics in the periphery, consistent with reduced activity in the core and greater activity in the viable rim. Doxorubicin treatment altered both surface morphology and dynamic pattern, and the region beneath a strongly scattering core remained observable in SC-FFOCM whereas it became dark in a point-scanning comparison because of attenuation and confocal loss (Tateno et al., 13 Nov 2025).

High-NA deep-tissue d-FF-OCM extended the application range to fresh ex vivo mouse liver and small intestine. In liver, dynamic contrast revealed hepatocyte borders, intracellular filamentous low-frequency networks interpreted as likely related to mitochondrial dynamics, nuclear activity, sinusoids, erythrocytes, and cells in sinusoids interpreted as likely immune or blood elements. In intestine, mucosal-side imaging revealed villi, an epithelial mosaic, nuclei, microvilli-like structures, probable goblet cells, and vascular networks, while serosal-side imaging showed myenteric plexus, submucosal plexus, muscle layers, crypts, and Paneth cells (Tarvydas et al., 5 Aug 2025).

6. Limitations, variants, and open problems

Despite its functional contrast, d-FF-OCM remains constrained by motion sensitivity, penetration, and interpretation. Foundational D-FFOCT was described as highly motion sensitive and mainly applicable to fresh ex vivo imaging, with typical penetration depths of $111$9–$n=1.4$0. In swept-source full-field OCM, the absence of a confocal gate avoids confocal signal loss with defocus, but also makes the method more vulnerable to stray-light artefacts, coherent crosstalk, and multiple scattering; the authors explicitly identified multiple scattering as a serious issue in deeper regions and noted that tissue-induced defocus and aberrations may vary laterally (Apelian et al., 2016, Tateno et al., 13 Nov 2025).

Specific limitations have produced corresponding hardware variants. iSR D-FFOCT addresses vibrations and fringe artefacts near reflective coverslips by using the coverslip itself as a self-reference. At more strongly reflective interfaces, dynamic dark-field FFOCT (D-dFFOCT) suppresses dominant substrate reflection by selective detection of scattered light, while dynamic reflection differential phase contrast (D-RDPC) uses asymmetric illumination to generate a directional dynamic contrast that reverses under illumination inversion and whose spatial localization can be recovered by directional Hilbert-transform reconstruction (Monfort et al., 2023, Monfort, 19 Jun 2026).

The broader trajectory of the field is toward deeper imaging, more stable interferometric geometries, and faster computation. Rolling-phase acquisition shows that static and dynamic contrast can be recovered from the same stack while reducing phase-offset bias (Monfort et al., 14 Jan 2025). High-NA deep-tissue systems show that broadband, bright, incoherent illumination and real-time reference-arm adjustment can extend dynamic imaging in highly scattering tissue to approximately $n=1.4$1 depth (Tarvydas et al., 5 Aug 2025). Learned four-volume LIV reconstruction reduces data handling substantially, but it remains tied to the conventional $n=1.4$2-volume reference definition. Collectively, these developments suggest that the central open problem is no longer whether full-field coherence microscopy can generate endogenous dynamic contrast, but how to do so with controlled artefacts, interpretable temporal metrics, and practical throughput across thick, heterogeneous, and clinically relevant specimens (Komeda et al., 15 Jan 2026).