math/aaa
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3434
Spectroscopic correlation tomography (SpCT)
for visualization of spatial correlations in
volumetric OCT scans
ERINC. O’KANE,
1,*
ROBERTE. HIGHLANDIII,
1
VENKATARAMANA THIRIVEEDI,
2
LAURENE. PARKER,
3
ALEKSANDRA
TATA,
4
PURUSHOTHAMA RAOTATA,
5
RAVIKARRA,
3,6
JATIN
ROPER,
2
DAVIDA. MILLER,
1
ANDADAMWAX
1
1
Department of Biomedical Engineering, Duke University, Durham, NC 27707, USA
2
Division of Gastroenterology, Department of Medicine, Duke University School of Medicine, Durham, NC
27710, USA
3
Division of Cardiology, Department of Medicine, Duke University Medical Center, Durham, NC 27707,
USA
4
Department of Surgery, Surgical Sciences, Duke University School of Medicine, Durham, NC 27710, USA
5
Departments of Cell Biology and Medicine, Duke University School of Medicine, Durham, NC 27707, USA
6
Department of Pathology, Duke University School of Medicine, NC 27707, USA
*
erin.okane@duke.edu
Abstract:
Spectroscopic optical coherence tomography (SOCT) allows for the targeted analysis
of cell nuclei due to their spectrally distinct scattering properties. SOCT has been explored for
optical biopsy based on its ability to extract structural and spectroscopic properties of a sample
without the need for labels. Using the dual-window method for SOCT, spatial correlations can be
computed throughout volumetric scans and analyzed to determine the size of cell nuclei. Such
analysis could enable quantitative visualization of tissue boundaries that appear homogenous
in structural OCT images. Here, we present spectroscopic correlation tomography (SpCT), a
technique for mapping spatial correlation measurements to a hue-saturation-value (HSV) color
space to help visualize relative changes in spatial correlation throughout volumetric OCT samples.
SpCT B-Scans and volumes of polystyrene bead samples with diameters 5.1µm to 11.3µm
were acquired and scatterer sizes were quantified using a Gaussian mixture model, achieving
size estimates within half-wavelength accuracy. 3T3 fibroblasts were imaged and analyzed via
connected component analysis, producing strong agreement with scatterer size measured via
confocal microscopy, thus validating our approach. Histology versus SpCT B-Scan comparisons
were performed for human colon organoids, cardiac organoids, and human lung airway explant to
demonstrate this technique’s ability to identify varying degrees of nuclear morphology in different
regions. SpCT volumes were used to objectively discriminate between control andAPC-mutated
human colon organoids based on a connected component analysis, identifying regions of long
spatial correlation and computing the coefficient of variation of the region volumes, which points
to a potential clinical use of this technique.
© 2026 Optica Publishing Group under the terms of the
1. Introduction
Optical coherence tomography (OCT) is a label free, interferometric imaging technique that
enables high-resolution cross-sectional imaging of biological samples [1,2]. Spectroscopic OCT
(SOCT) is a functional extension of OCT that decomposes the broadband OCT signal into
spectral sub-components, using the short-time Fourier transform (STFT) where the spectrum is
filtered by a sliding window before reconstruction [3–6]. However, the window width introduces
a tradeoff between spectral and axial resolution: broader windows improve axial resolution, while
#593432 https://doi.org/10.1364/BOE.593432
Journal © 2026 Received 13 Feb 2026; revised 15 May 2026; accepted 19 May 2026; published 3 Jun 2026
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3435
narrower windows enhance spectral resolution. To overcome this tradeoff, we developed the
dual-window (DW) method, which computes the STFT using both broad and narrow windows
and multiplies the resulting time-frequency distributions (TFDs) [7]. A distinctive feature of the
DW TFD is the presence of local spectral oscillations around each scatterer, whose frequencies
encode scatterer size. Fourier analysis of these oscillations yields the correlation distance (CD)
function, representing spatial correlations within and between scatterers [7–10].
Previously, we demonstrated accurate, high-throughput size measurement of polystyrene
spheres ranging from 5 to 11µm in diameter using the CD function generated from the DW-SOCT
method [10]. The technique accurately discriminated the size of strong scatterers, however, for
weaker scatterers, limitations arise when samples have lower backscattering signal (due to a
lower relative refractive index). Further, this approach was limited to isolated scatterers, such as
suspended beads or cells. Therefore, to effectively translate spectroscopic correlation analysis to
more complex samples, such as tissues, there is a need for a continuous, pixel-wise approach that
enables visualization of spatial correlations throughout an OCT volume.
For this work, we conduct OCT imaging of organoids, miniaturizedin vitroorgans or tissues
that mimic the function and morphology of a particular organ [11,12]. They are cultured by
extracting stem cells fromin vivotissues, which self-organize and differentiate into specialized
cell types. These 3D cell cultures have major applications in clinical research, such as disease
modeling and drug delivery. Organoids enable systematic analysis via gene editing, which is
particularly useful to study diseases difficult to modelin vitroand serve as an alternative to animal
models. For example, current efforts in colorectal cancer research have used patient-derived
colon organoids to study personalized drug development, and early colorectal cancer detection
[13,14].
Cell nuclear size is a strong biomarker for dysplastic tissue, as benign epithelial cells range from
5µm – 10µm in diameter, while cancerous epithelial cells enlarge up to 20µm [15,16]. The larger
refractive index difference of cell nuclei in comparison to other cell and tissue structures allow
for its specific analysis using techniques such as SOCT [8,17]. The spectroscopic analysis and
measurement of nuclear sizes throughout human derived organoids can aid in the development
of an objective and robust differentiation of healthy and abnormal tissues. Developing these
methods using human-derived organoids will aid in the translation toex vivoandin vivoanalysis.
In this paper, we introduce a method to effectively visualize spatial correlations throughout
a volumetric OCT scan, known as spectroscopic correlation tomography (SpCT), enabling
both quantitative and qualitative analysis of scatterers and their spatial distributions. First, we
introduce a hue-saturation-value (HSV) color space to visualize CD functions and validate the
color space using polystyrene beads suspended in collagen hydrogel andin vitrosamples of
3T3 fibroblasts. Next, we perform SpCT in various tissue samples including colonic organoids
mutated with the adenomatous polyposis coli (APC) gene, cardiac organoids treated with vascular
endothelial growth factor (VEGF), and lung airway explant. Lastly, we discriminate control and
APC-knockout colonic organoids by measuring the volumes of groups of long spatial correlations
and computing the coefficient of variation.
2. Methods
A conceptual schematic for the spectroscopic processing pipeline used to generate and analyze
SpCT images is shown in Fig.. Image acquisition is first performed using a balanced-detection
SOCT engine, followed by the dual window processing to generate the correlation distance (CD)
function from the spectroscopic data. Parameters derived from the CD function and structural
OCT data are mapped to the hue, saturation, and value channels, encoding spatial correlation
information within an HSV color space to produce volumetric SpCT images. Finally, statistical
analysis of the hue channel, including histogram-based Gaussian mixture modeling and connected
component analysis, enables quantitative measurement of scatterer size in polystyrene beads and
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3436
3T3 cell nuclei, as well as discrimination betweenAPC-knockout and control colon organoids.
Each spectroscopic processing pipeline step is detailed in the Methods section.
Fig. 1.
Spectroscopic OCT processing pipeline for SpCT image generation and analysis.
SOCT data are processed using a dual-window method to compute the correlation distance
(CD) function and scatterer size, which are encoded with structural OCT intensity in an
HSV color space to generate volumetric images. Hue-based statistical analysis enables
quantitative size measurement and classification.
2.1. Experimental setup
Data was collected using a custom built balanced-detection (BD) OCT engine described previously
[10]. Light from a supercontinuum laser (SuperK Extreme EXW-4, NKT Photonics) was filtered
to deliver a 150 nm bandwidth centered at 900 nm and launched to a 70:30 2×2 fiber coupler.
The 30% output illuminated the sample arm comprised of a fiber collimator, MEMS scanning
mirror, relay optics, and a 0.25 NA, 10x objective. The 70% output illuminated the transmission
reference arm that consisted of a fiber collimator input, retroreflector mounted on a linear
translation stage, and a fiber collimator output. Backscattered light from the sample arm and
transmitted light from the reference arm recombined at a 50:50 2×2 fiber coupler that splits light
between two spectrometers. This second fiber coupler introduces aπphase shift between the
recorded interference fringes enabling suppression of common noise, such as relative intensity
noise (RIN), and doubling of the signal amplitude when the simultaneously captured fringes are
subtracted [18–20]. The spectral range of the combined spectrometers is 798 nm – 1000 nm, with
a 6 dB sensitivity roll-off at 1.11 mm. Volumetric OCT scans are acquired at 40 kHz with 400
A-lines per B-Scan, and 512 B-Scans per volumetric scan. The axial resolution is 1.8µm and the
lateral resolution is 5µm.
2.2. SOCT processing
Automated analysis of the CD function at each point in a volumetric OCT scan allows for in-depth
analysis and detailed visualization of spatial correlations throughout a sample [10]. First, standard
BD-SD-OCT processing is performed on each spectral interferogram including balancing fringes
[20], wavenumber linearization, and dispersion compensation [21]. The DW method is applied to
each A-line, involving the STFT of both broad and narrow windows individually, then multiplying
the resulting spectra. Broad and narrow windows had bandwidths of 0.0167cm
−1and 0.333
cm
−1
, respectively, corresponding to coherence lengths of 200µm and 10µm.
The STFT generates the TFD, where each depth has a corresponding spectrum [3,10]. At
each depth, residual spectral bias is removed by subtracting the spectrum’s 1D Gaussian fit.
The Fourier transform is taken of each spectrum, yielding the CD function. Scatterer-induced
oscillations in the spectra have a periodicity guided by the scatterer’s round-trip optical path
length,∆OPL, allowing for the computation of scatterer size,dc, for a given refractive index of
the scatterer,ns, and relative refractive index,nrel:dc=∆OPL/ 2nsnrel[9,10]. Prominent peaks
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3437
in the CD function indicate strong spatial correlations, which includes the scatterer’s diameter.
The CD function is generated for each point in a B-Scan or volumetric OCT scan.
2.3. Definition of the hue-saturation-value (HSV) color space
The HSV color space provides an intuitive representation of colors in an RGB color model. To
effectively visualize spatial CDs in a 2D or 3D image, parameters of the structural OCT B-Scan
and CD function are assigned to hue, saturation, and value channels of an HSV color space,
shown in Fig.. The hue channel encodes the location of the CD function’s maxima (Fig.(a)),
the saturation channel encodes the intensity of the CD function’s maxima (Fig.(a)), and the
value channel encodes the amplitude of the structural OCT B-Scan at each pixel (Fig.(b)). Hue,
saturation, and value channels are combined to generate the HSV image, which represents spatial
correlations as color and emphasizes scatterers (Fig.(c)).
Fig. 2.
Illustration of HSV color space definition with (a) hue and saturation represented by
the maximum CD peak location and intensity, respectively, (b) value represented by OCT B-
Scan intensity, and (c) the combination of H, S, and V channels to generate spectroscopically
encoded colorful image, emphasizing strong scatterer
2.4. Sample preparation
2.4.1. Polystyrene beads in collagen hydrogel
Polystyrene beads (NIST Traceable Standards, n=1.5755 atλ=850 nm) with diameters of
5.1±0.3µm, 7.7±0.4µm, and 11.3±0.3µm were suspended at various depths in methacrylated
collagen hydrogel (PhotoCol with LAP Kit, Advanced Biomatrix, n=1.33 [22]). Collagen
hydrogel was reconstituted with 20 mM acetic acid at 4 mg/mL. For each mL of collagen, 90µL
of neutralization buffer (Advanced BioMatrix) and 0.1µg photoinitiator were added. Polystyrene
beads were added and distributed thoroughly by mixing via vortex for 30 seconds. Once sample
was transferred to sample holder, photocrosslinking was initiated with 365 nm ultraviolet light
source for up to one minute.
Two-size polystyrene bead samples were manufactured by first preparing size A in an individual
sample holder and photocrosslinking with UV source. A punch biopsy tool removed a 1mm
3
cylinder from size A sample. Size B was prepared and transferred into removed cylinder of size
A, the photocrosslinked with UV source.
2.4.2. 3T3 fibroblasts in DMEM
3T3 mouse embryonic fibroblast cells were cultured in Dulbecco’s Modified Eagles Medium
(DMEM) (Thermofisher) supplemented with 10% fetal bovine serum (Avantor) and 1% Pen-Strep
antibiotic (Corning) in a 5% CO237°C environment. Every 2-3 days the cell line was passaged
using 0.25% trypsin-EDTA (Gibco). To prepare for SpCT imaging, 3T3 cells were seeded
with media onto agar-coated 8-chamber microslides (Ibidi). Agar was prepared by dissolving
bacteriological agar powder (Sigma) in distilled water, heating the solution to boiling, and using
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3438
it to coat the microslides. For nuclear staining and confocal imaging, 3T3 cells were suspended
in media and stained with DAPI (Invitrogen). Following a brief incubation, cells were mixed
with methacrylate-conjugated bovine Type I collagen (Advanced BioMatrix) and polymerized
in an 8-chamber microslide (Ibidi). Prior to cell incorporation, lyophilized type I collagen was
dissolved in 20 mM acetic acid at 4 mg/mL according to manufacturer’s protocol. Confocal
imaging was performed using a Zeiss LSM 880 laser scanning confocal microscope with a 20X
objective, acquiring 20 z-stacks with a step size of 0.68µm.
2.4.3. Colon organoids
Colon organoids were generated from endoscopic human biopsy samples from Duke University
Hospital under Duke IRB approval (00104775). Organoids were cultured in 3D Matrigel with
L-WRN conditioned media [23], appropriately supplemented, and contained in 96-well plate.
TargetedAPCmutations were introduced using CRISPR-Cas9 sgRNA genes. By incubating
sgRNA with Cas9 nuclease,APC-targeting ribonucleoprotein (RNP) complexes were generated.
Colon organoids were enzymatically dissociated, and fragments were electroporated, recovered,
and re-embedded in Matrigel for a 7-day culture.APC-mutant clones were enriched with
Wnt/R-spondin-depleted conditions, screened with SURVEYOR assay, and validated by sequence
analysis. Colon organoids were prepared for histology using an optimized FFPE protocol
[24]. Organoids were fixed in formalin, dehydrated through graded ethanol, cleared in xylene,
and infiltrated with paraffin wax at 60°C for 2 hours before sectioning at 4µm. Sections
were deparaffinized, rehydrated, stained with H&E, and imaged using a Leica DMi8 inverted
microscope. Detailed procedure for colon organoid preparation is presented in
(see Table S1).
2.4.4. Cardiac organoids
Cardiac organoids were prepared from induced pluripotent stem cells (iPSCs) from a healthy
male donor [25]. Cells were differentiated into cardiac mesoderm using a standard Wnt
activation/inhibition protocol, followed by replating in suspension culture in an ultra-low
attachment plate to induce organoid formation [26,27]. Endothelial cell differentiation was
induced by supplementing media with 50 ng/mL VEGF-165. For OCT imaging, organoids were
fixed in 4% paraformaldehyde, embedded in low-melt agarose, and solidified before imaging. For
staining and confocal microscopy, organoids were fixed in acetone, incubated with primary and
secondary antibodies (anti-human CD31 and Alexa-Flour 488), as well as DAPI, and mounted
on coverslip for imaging [28]. Confocal imaging was performed on Zeiss AxioObserver Z1
Spinning Disk confocal microscope with 40X objective and 1.95µm z-stacks. Detailed procedure
for cardiac organoid preparation is presented in Supplement 2 (see Table S2).
2.4.5. Lung explanted tissue
Healthy donor lungs were obtained in accordance with Institutional Review Board oversight (Duke
University Pro00114526 – “Human Lung Stem Cells”; exempt research as described in 45 CFR
46.102(f), 21 CFR 56.102(e) and 21 CFR 812.3(p) which satisfies the Privacy Rule as described in
45CFR164.514). Human large airway segments were dissected and cut into small explant pieces
under sterile conditions. Porous Gelfoam sponges were prepared to support explant culture at an
air-liquid interface(ALI). Gelfoam was cut into uniform pieces, saturated in DMEM/F12 medium
supplemented with 10% fetal bovine serum (FBS), 1% penicillin-streptomycin-amphotericin B,
and 50µg/mL gentamicin, and transferred into individual wells of a tissue culture dish. Airway
explants were placed on top of the media-soaked Gelfoam with the luminal surface facing upward.
Additional culture medium was added to the bottom of each well to maintain hydration of the
Gelfoam without submerging the explant, thereby preserving the ALI. Cultures were maintained
at 37°C with 5% CO2, and media was replaced every 2–3 days.
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3439
2.5. Statistical analysis
Scatterer size measurements of polystyrene beads from the SpCT images were obtained by
analyzing the hue channel histogram of each single B-Scan. Because the hue channel is assigned
to the length scale of the CD function maxima, the resulting histogram contains contributions from
both true scatterer diameter and multiple-scattering peaks. To extract the true scatterer diameter,
we analyzed the modality (number of peaks) of each hue histogram and fit the distribution using a
Gaussian mixture model (GMM). The mean and standard deviation of each identified peak were
then computed. Based on our prior multiple scattering analysis [10], the peaks corresponding to
the scatterer diameters were identified and only these were considered for the analysis here. To
isolate the CD of 11.3µm beads, histogram peaks within the 8µm to 14µm range were considered.
For 7.7µm beads, histogram peaks between 6µm to 9µm were selected, and for 5.1µm the range
was 4µm to 7µm. Peaks outside these ranges were attributed to multiple-scattering contributions
and excluded from further analysis. Overlap between histogram peaks and distributions were
expected due to the inherent size distribution of each bead population. Therefore, the GMM
was applied to capture each scatterer size’s full distribution while minimizing interference from
adjacent CD peaks.
Nuclear size in 3T3 cells was quantified from SpCT images using a connected component
analysis. Regions exhibiting relatively long spatial correlation spanning at least 10 connected
voxels were identified, ensuring that only structures above the CD background signal were included.
The minimum size threshold (≥10 voxels) was determined based on geometric considerations
and imaging resolution. With an effective pixel length of∼1.5µm in hydrogel, a nucleus at the
lower bound of the expected size range (∼8µm diameter, based on confocal measurements) would
span∼4-5 pixels axially, and∼1-2 pixels laterally at the center, corresponding to a minimum
of∼10 connected voxels. Spatial correlation values less than 7µm were excluded, as they are
consistent with scattering from other structures. This threshold was selected based on confocal
microscopy measurements of 3T3 fibroblast nuclear size (9.5±1.5µm, N=30), corresponding
to a size range of∼8-11µm. The lower bound was extended to 7µm to account for variability
and ensure inclusion of smaller nuclei within the distribution tail. For each connected region, the
mean CD value was reported as the corresponding cell’s nuclear size. The distribution of nuclear
sizes across all detected regions was summarized by reporting the mean and standard deviation
of the sample. A SpCT-measured nuclear size within±λ/2 of a confocal-microscopy measured
size is deemed an accurate measurement.
For colonic organoids, connected component analysis quantitatively compares control and
APC-knockout samples. We identify regions of strong scatterers with spatial correlations
greater than 6µm, corresponding to cell nuclei. This threshold was determined based on the
observed separation between background and nuclear signals in the spectroscopic contrast, where
background regions predominantly exhibit correlations below 6µm, and nuclei spear as circular
regions with larger correlation lengths. These regions are segmented in 3D SpCT volumes, and
their individual volumes are measured. For each SpCT image, we compute the mean and standard
deviation of the segmented region volumes and use these values to calculate the coefficient of
variation (CV), defined as the ratio of the standard deviation to the mean. CVs from all SpCT
images within each group (control andAPC-knockout) are then averaged, with the standard
deviation reported as uncertainty. Statistical significance between groups is assessed using a
Mann-Whitney U test, with p<0.01 indicating a statistically significant difference in spatial
correlation distributions between control andAPC-knockout groups.
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3440
3. Results
3.1. SpCT validation
We validated the SpCT visualization approach using different combinations of polystyrene
beads suspended in collagen hydrogel, andin vitro3T3 cell samples. Figure
resulting SpCT B-scans for 11.3µm polystyrene beads suspended in collagen hydrogel. The hue
channel (Fig.(a)) encodes the spatial correlation, the saturation channel (Fig.(b)) encodes the
correlation amplitude, and the value channel (Fig.(c)) encodes the full-band OCT amplitude.
The combined HSV image (Fig.(d)) provides a colorized visualization of the scatterer size
scaled by the OCT signal amplitude. The hue of the beads in the combined image provides an
accurate estimate of bead diameter as indicated by the black arrow near the color bar of Fig.(d).
The average scatterer size is measured to be 11.9±0.9µm (N=8347 pixels), which is within
±λ/2 of the theoretical bead size, indicating an accurate measurement. Note that a range of hue
values span the depth of each bead, which we attribute to multiple scattering events causing
longer spatial correlations as the result of a high relative refractive index and larger scatterer size
[10]. This is further addressed in the Discussion section.
Fig. 3.
Correlation tomograph channels of (a) hue, (b) saturation, (c) value, and (d) H, S,
and V combined of 11.3µm polystyrene beads suspended in collagen hydrogel, with (e)
zoomed-in section. Scale bar=100µm.
Figure
hydrogel, demonstrating clear separation between scatterers with distinct diameters. The first
sample containing 11.3µm and 5.1µm beads, separated by a defined boundary, is shown through
a SpCT B-Scan (Fig.(a)), en face slice (Fig.(b)), and volumetric scan (Fig.(c)). This sample
yielded measured diameters of 11.4±1.0µm (N=1673 pixels) and 5.7±1.1µm (N=1748
pixels), for the two populations of bead sizes respectively. A second sample comprised of 11.3µm
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3441
and 7.7µm beads shows a distinct margin of separation, also visualized using a SpCT B-Scan
(Fig.(d)), en face slice (Fig.(e)), and volumetric scan (Fig.(f)). Measured diameters for this
sample were 11.5±0.9µm (N=1548 pixels) and 7.7±1.2µm (N=1753 pixels), respectively for
the two populations of bead sizes. Average scatterer sizes were all within±λ/2 of the expected
size, indicating accurate measurements. Supplement 3 includes videos of volumetric SpCT
images for each two-size sample (seeVisualization 1 ).
Fig. 4.
SpCT images of two-channel polystyrene samples. (a) B-Scan projection of 11.3 um
beads (left) and 5.1 um beads (right), (b) en face projection of 11.3 um beads (top) and 5.1
um beads (bottom), (c) volumetric scan of 11.3µm and 5.1µm beads (seeVisualization 1),
(d) B-Scan projection of 7.7 um beads (left) and 11.3 um beads (right), (e) en face projection
of 7.7 um beads (top) and 11.3 um beads (bottom), and (f) volumetric scan of 7.7µm and
11.3µm beads (seeVisualization 2). Scale bar=100µm.
SpCT images of 3T3 cells on a layer of bacteriological agar are presented in Fig.. The
individual hue (Fig.(a)), saturation (Fig.(b)), and value (Fig.(c)) channels, as well as a
B-Scan slice of the combined HSV SpCT image (Fig.(d)) are shown. A histogram of the
hue channel values (Fig.(e)) shows the distribution of spatial correlations, as well as the hue
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3442
scaling emphasizing correlations from 4µm to 12µm. The SpCT en face projection (Fig.(g))
demonstrates circular nuclei, producing measured sizes from 8µm to 11µm in diameter. Zoomed-
in regions of the B-Scan (Fig. (5(h)) and enface (Fig. (5(i)) slices enable detailed visualization
of individual circular nuclei. To obtain an objective measurement of cell nuclei via the SpCT
image, a connected component analysis is performed. Regions of spatial correlation above the
background CD (CD>7µm), with at least 10 connected voxels were identified as individual cells,
and the average CD within each cell reported as its nuclear size. The histogram of measured
nuclear sizes is presented (Fig.(f)), with an average nuclear size of 9.0 ±0.8µm (N=480).
Confocal microscopy validated the nuclear size, yielding an average diameter of 9.5±1.5µm
(N=30). The SpCT-measured nuclear size within±λ/2 of the confocal-measurement validates
the accuracy of this approach. Supplement 4 presents video flythroughs and a volumetric
projection of the 3T3 fibroblast sample (seeVisualization 3 ). Supplement 5 presents the en
face projections of the hue, saturation, and value channels of the 3T3 fibroblasts (seeFigure S1).
Fig. 5.
SpCT images of 3T3 fibroblasts layer on bacteriological agar. B-Scan (zx) projection
of (a) hue, (b) saturation, and (c) value channels, and (d) combined hue-saturation-value
(HSV) image (seeVisualization 3). (e) Histogram of hue channel with applied color bar
scale. (f) Histogram of mean voxel CD of connected regions (g) En face (xy) projection
HSV SpCT image (seeVisualization 4). (h) Zoomed-in section from B-Scan projection. (i)
Zoomed-in section from enface projection. Scale bar=100µm.
3.2. Correlation tomography in organoids and tissue
We performed SpCT on colonic and cardiac organoids as well as explanted lung tissue. An
en face projection of anAPCedited human colonic organoid OCT image (Fig.(a)) and its
corresponding SpCT image (Fig.(b)) are presented. Representative histology of APC-mutated
colon organoids with hematoxylin and eosin (H&E) staining is presented (Fig.(c)). The
APC-knockout suppresses theAPCgene responsible for tumor suppression, inducing aneuploidy
and used to model advanced cancers; therefore, regions of greater cell density and nuclear size
due to increased proliferation, corresponding to longer spatial correlations are expected. The
structural OCT image (Fig.(a)) reveals some regions with higher backscattering intensity but
lacks fine detail and quantitative metrics that could be used to discriminate organoid type. In
contrast, the SpCT image (Fig.(b)) emphasizes regions of long and short spatial correlations
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3443
surrounding the individual organoids, with clusters of long spatial correlations measuring up to
8µm. These spatial patterns are more prominent in the SpCT image than the structural image and
well align with the confocal image (Fig.(c)), which shows EdU (indicating cell proliferation)
regions on the outer edge.
Fig. 6.
Colon organoid structural and spectroscopic images compared to histology.APC-
knockout colon organoid (a) OCT en face projection, (b) corresponding SpCT image, and
(c) representative H&E histology. Control colon organoid (d) OCT en face projection, (e)
corresponding SpCT image, and (f) representative H&E histology. Scale bar 100µm.
Figure
(Fig.(a)), its corresponding SpCT image (Fig.(b)), and representative confocal microscopy
image of VEGF-treated (Fig.(c)) and control (Fig.(d)) organoids, with stained endothelial cells
(CD31, yellow), and nuclei (DAPI, blue). VEGF-treatment promotes endothelial cell growth;
therefore, regions of long spatial correlations associated with high cell proliferation are expected.
The structural OCT image (Fig.(a)) shows a bright region but lacks detail in structure and
quantification. The SpCT image (Fig.(b)) displays the same bright upper region but also enables
the visualization of long spatial correlations within the dense scatterer region. This is strongly
supported by the confocal image (Fig.(c)), which shows a confined region of endothelial cell
presence, consistent with the VEGF-induced proliferation, resulting in larger nuclear sizes and
dense cellular regions compared to the control organoid.
Figure(a)),
its corresponding SpCT image (Fig.(b)), and representative histology of pseudostratified
ciliated columnar epithelium with H&E staining [29] (Fig.(c)). The structural OCT image
(Fig.(a)) demonstrates a strong scattering superficial layer. The SpCT image (Fig.(b)) enables
the identification of long correlations within this layer, and a less dense layer of cells beneath.
Histology of pseudostratified ciliated columnar epithelium (Fig.(c)) confirms a dense layer of
cells near the surface with a longitudinal orientation (long-axis vertical), and a sparser distribution
of cells at greater depths with smaller nuclei, oriented laterally (long-axis horizontal). The SpCT
image is consistent with the histology, showing well-defined spatial correlations up to 12µm, and
deeper regions containing less dense and shorter spatial correlations.
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3444
Fig. 7.
(a) OCT en face projection of VEGF-treated human cardiac organoid, (b) correspond-
ing spectroscopic correlation tomograph, and (c) representative histology of VEGF-treated
cardiac organoid and (d) control cardiac organoid via confocal microscopy highlighting
endothelial cells (CD31, yellow), and nuclei (DAPI, blue). Scale bar 100µm.
Fig. 8.
(a) OCT B-Scan image of human lung airway explant, (b) corresponding spectro-
scopic correlation tomograph, and (c) representative histology via brightfield imaging with
H&E staining. Scale bar=100µm.
3.3. Quantitative classification of colon organoids using correlation tomography
To demonstrate the quantitative analysis power of SpCT, we developed a method for classifying
between control andAPC-treated colon organoids based on spatial correlation values. Volumetric
SpCT images are generated for eachAPC-knockout (Fig.(a)) and control (Fig.(c)) colon
organoid sample. Example B-Scan slices within each volumetric SpCT image for theAPC-
knockout (Fig.(b)) and control (Fig.(d)) are presented. For each of the N =20APC-knockout
and N=20 control colon organoid samples, regions with OCT intensity above 65% the maximum
and spatial correlations greater than 6µm were isolated in the SpCT volumes. These regions
of long-correlation scatterers, which likely correspond to individual nuclei or nuclei clusters,
were segmented and their volumes computed. The distribution of these volumes for each sample
was visualized as a histogram (Fig.(e)), and the mean, standard deviation, and coefficient of
variation (CV) computed. CVs were then averaged across samples within each group, with
the standard deviation reported as uncertainty. Control organoids had a mean CV 1.5±0.3,
whileAPC-knockout organoids had a larger mean CV of 1.7±0.4. A Mann Whitney U-test
revealed statistical significance between results (p=0.007), indicating a significant difference in
cell nuclei organization inAPC-knockout organoids (Fig.(f)).
4. Discussion
SpCT offers enhanced visualization of scatterer spatial organization and proliferative activity
across variousin vitrosamples, including polystyrene beads, 3T3 cells, human colon and cardiac
organoids, and explanted lung tissue. Foundational light-scattering studies have demonstrated
that variations in angular and spectral scattering signatures can be used to quantify scatterer size
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3445
Fig. 9.
Quantitative discrimination ofAPC-treated and control human colon organoids.
(a) Volumetric and (b) B-Scan SpCT images ofAPC-knockout colon organoids. (c)
Volumetric and (d) B-Scan SpCT images of control colon organoids. (e) Corresponding
histogram of long-CD region volumes forAPC-knockout (red: 3.1±7.3 pixels, CV=2.3)
and control (blue: 2.5±2.9 pixels, CV=1.2) organoids. (f) Mean coefficient of variation
forAPC-knockout and control organoids (p<0.01). Scale bar=100µm.
and cellular morphology in biological media [30]. Building on this, depth-resolved spectroscopic
techniques such as Fourier-domain low-coherence interferometry enable extraction of scattering
information with subwavelength sensitivity to structural features in subsurface layers [31]. The
approach was demonstrated to provide accurate nuclear size measurements through comparison
with confocal microscopy measurements [32]. Previous development as a functional imaging
method includes the selective measurement of polystyrene bead sizes using the DW SOCT
method, assigning measurements to a singular hue, and overlaying colors onto an OCT B-Scan
[9]. We build upon these foundational methods by automating the scatterer selection process
and performing a rigorous analysis on the CD function, identifying the signatures of multiple-
scattering events [10]. Other methods for scatterer size measurement and visualization include the
correlation of the derivative method, and overlaying hue-encoded size estimates onto structural
OCT images [17,33]. This approach has strengths in being a pixel-wise approach that includes
a wide range of hues, allowing for variations in scatterer size throughout depth to be carefully
observed. We advance this method by removing scatterer size overlays of the OCT image and
instead performing spectroscopic analysis at every point with B-Scan intensity above the noise
floor. To enhance visualization, we further seek to improve contrast between strong and weak
scatterers by implementing a dynamic saturation channel, dictated by the scattering intensity.
Our proposed pixel-wise approach with an HSV color space places emphasis on strong scatterers
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3446
and enables spatial pattern visualization throughout a volumetric scan, allowing for thorough
quantitative and qualitative spectroscopic analysis.
SpCT imaging of polystyrene beads enabled visualization of scatterer sizes throughout 2D
and 3D scans, as well as rapid quantification of scatterer sizes in a 2D image. Scatterer size was
encoded in the hue channel, allowing intuitive visualization and straightforward measurement
via hue histogram analysis, with size estimates accurate to within half a wavelength. However,
accuracy and precision of this technique were limited when only a single cross-section was
analyzed, as the identified CD could underestimate the true scatterer size when the B-Scan
did not intersect the bead’s center. Our previous work accurately measured polystyrene bead
diameters throughout a volumetric scan by isolating the center of each bead in 3D space and
computing the CD function at that location [10]. This peak-finding approach improved accuracy
and addressed multiple scattering artifacts, particularly in samples with large scatterers and
high relative refractive index [10]. We found that the CD peak corresponding to the true
scatterer size was not necessarily the most intense or prominent peak, therefore requiring careful
identification. In SpCT, the hue channel is assigned to the maximum of the CD function at each
pixel, which may reflect a multiple-scattering peak rather than the true diameter. To mitigate
this in polystyrene bead samples, our current method uses a Gaussian-mixture model to match
hue histogram distributions to each bead size’s theoretical CD distribution, guided by our prior
multiple-scattering characterization [10]. This approach significantly reduced computational
time, while preserving reliable scatterer size measurements.
The effective visualization and accurate measurement of cell nuclei was demonstrated using
SpCT imaging of 3T3 fibroblasts. Analysis of the individual hue, saturation and value channels
confirmed that the spectroscopic signal originated from the cells. Cell nuclei were clearly
emphasized in both SpCT B-Scan and en face images, with a measured nuclear diameter via
connected component analysis of 9.0±0.8µm, consistent within±λ/2 of the confocal microscopy
measurement of 9.5±1.5µm.
SpCT images provided enhanced visualization of spatial correlation organization and pro-
liferative activity across human colon and cardiac organoids, as well as explanted lung tissue,
compared to structural OCT alone. InAPC-knockout human colon organoids, regions of long
spatial correlations surrounding each organoid aligned with the EdU-labeled proliferative regions.
The VEGF-treated cardiac organoids demonstrated a dense region of long spatial correlations,
corresponding to endothelial cell proliferation and enlarged nuclei, confirmed with the confocal
microscopy. In the human lung tissue, the correlation tomograph displayed a superficial layer of
strong scattering and relatively long correlations, followed by a deeper, less dense region, which
is consistent with histological features of pseudostratified ciliated columnar epithelium. Together,
these results highlight the utility of SpCT in resolving microstructural and proliferative features
across diverse tissue types.
A quantitative method for distinguishing control andAPCedited colon organoids by analyzing
cell nuclei organization was demonstrated by analyzing regions of long spatial correlations in
SpCT volumes via connected component analysis. Measuring regions with spatial correlations
greater than 6µm, which likely correspond to individual or clusters of nuclei, revealed significantly
higher CVs inAPCdeleted samples. These findings are consistent with loss of theAPCgene,
including enlarged and clustered nuclei arising from aneuploidy and increased cell proliferation.
These results suggest that SpCT volumes can help differentiate normal from dysplastic tissue
based on spatial correlation organization. However, limitations of this analysis include variability
in organoid developmental timing, which may influence the stage of precancerous or cancerous
tissues and complicate interpretation. While the current approach establishes a promising
framework for spectroscopic discrimination of dysplasia, future studies will require longitudinal
monitoring of organoids to rigorously track the progression of dysplasia over time.
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3447
The analysis ofin vitrosamples, including cultured 3T3 cells, colon and cardiac organoids, and
lung explanted tissue, via SpCT shows promise in the quantitative and qualitative discrimination
of normal and dysplastic tissues. This technique is broadly applicable to a variety of tissues for the
label-free, volumetric assessment of nuclear sizes and distributions. The utility of SpCT depends
on the underlying scattering properties of the sample. In low-scattering tissues, the reduced
backscatter signal can lead to lower signal-to-noise ratios, and therefore weaker spatial correlation
peaks and limited accuracy in scatterer size estimation. In layered tissues, such as the lung explant
tissue, variations in refractive index and backscattering intensity across depth can introduce
depth-dependent correlations, which can reduce scatterer visibility and accuracy at deeper layers.
In highly scattering tissues, multiple-scattering contributions from adjacent strong scatterers
could produce spurious long correlation peaks, as well as reduce contrast between relevant
scatterers and background signal. This could complicate the analysis of individual scatterers
and reduce specificity. Mitigating these sources of error, including noise-limited CD estimation,
depth-dependent correlation effects, and overlap between single- and multiple-scattering signals,
would require appropriate filtering and statistical analysis.
Future applications of this work involve SpCT of humanex vivocolon tissue to develop objective
discrimination between normal and adenomatous colon polyps based on spatial correlations.
These classification methods via SpCT can be incorporated into a deep-learning classifier for
early CRC detection.
5. Conclusion
In this paper, we introduce SpCT as a powerful extension of SOCT for visualizing and quantifying
spatial correlations across 2D and 3D images. Mapping spectroscopic parameters to HSV color
space enables SpCT to provide intuitive and accurate visualization of scatterer size distributions.
This technique is validated through SpCT analysis of polystyrene bead and 3T3 fibroblast imaging,
demonstrating the intuitive visualization of scatterer sizes and accurate measurements via hue
histogram analysis. Application of SpCT to human colon and cardiac organoids, as well as lung
explanted tissue, demonstrated the ability to identify microstructural features and variations in
cell density, which were not apparent in structural OCT images. SpCT also enabled the objective
discrimination between control andAPC-knockout colon organoids based on the coefficient of
variation of long correlation regions. This highlights the potential of SpCT as a method for
early dysplasia detection. This work demonstrates the clinical utility of SpCT for label-free,
quantitative, and qualitative assessment of tissue compositions based on spatial correlations.
Future work will apply SpCT images to a deep-learning classifier to automatically discern normal
and dysplastic samples based on nuclear size. This work points the way towards the automated,
objective analysis to discriminate healthy, precancerous, and cancerous tissues.
Funding.
National Institutes of Health (R01AG072732, 1F30HL175908-01); U.S. National Science Foundation
(2125528); Duke University (John T. Chambers Scholars Award); National Heart Lung and Blood Institute (R01HL157277,
R01HL160939); American Heart Association (24PRE1186062); National Cancer Institute (5R37-CA259363-04, 1R01-
CA254108-04, 5R01-CA244359-05).
Acknowledgments.The authors thank Justin Fan for his experimental support.
Disclosures.A.W. is President of Lumedica Vision Inc. DAM is a consultant for Lumedica Vision Inc.
Data availability.
The data that support the findings of this study are available from the corresponding author upon
reasonable request.
Supplemental document.See
References
1.
J. A. Izatt and M. A. Choma, “Theory of Optical Coherence Tomography,” inBiological and Medical Physics,
Biomedical Engineering(Springer, 2008).
2.
J. G. Fujimoto, M. E. Brezinski, G. J. Tearney,et al., “Optical biopsy and imaging using optical coherence
tomography,” 1(9), 970–972 (1995).
Research Article Vol. 17, No. 7 / 1 Jul 2026 /Biomedical Optics Express3448
3.
R. N. Graf and A. Wax, “Temporal coherence and time-frequency distributions in spectroscopic optical coherence
tomography,” 24(8), 2186 (2007).
4.
W. Y. Kendall, Q. Tian, S. Zhao,et al., “Deep learning classification of ex vivo human colon tissues using spectroscopic
optical coherence tomography,” 17(9), e202400082 (2024).
5.
M. Wojtkowski, R. Leitgeb, A. F. Fercher,et al., “Spectral measurement of absorption by spectroscopic frequency-
domain optical coherence tomography,” 25(11), 820–822 (2000).
6.
J. Yi, X. Li, and J. Gong, “Analyzing absorption and scattering spectra of micro-scale structures with spectroscopic
optical coherence tomography,” 17(15), 13157–13167 (2009).
7.
F. Robles, R. N. Graf, and A. Wax, “Dual window method for processing spectroscopic optical coherence tomography
signals with simultaneously high spectral and temporal resolution,” 17(8), 6799–6812 (2009).
8.
F. E. Robles, Y. Zhu, J. Lee,et al., “Detection of early colorectal cancer development in the azoxymethane rat
carcinogenesis model with Fourier domain low coherence interferometry,” 1(2), 736 (2010).
9.
F. E. Robles and A. Wax, “Measuring morphological features using light-scattering spectroscopy and Fourier-domain
low-coherence interferometry,” 35(3), 360–362 (2010).
10.
E. O’Kane, W. Wang, R. I. Highland,et al., “Spectroscopic analysis of volumetric OCT data for automated
measurement of scatterer size,” 42, 1344–1353 (2025).
11.
S. Yang, H. Hu, H. Kung,et al., “Organoids: The current status and biomedical applications,” 4(3), 274
(2023).
12.
M. Almeqdadi, M. D. Mana, J. Roper,et al., “Gut organoids: mini-tissues in culture to study intestinal physiology
and disease,” 317, C405–C419 (2019).
13.
C. Yang, W. Xiao, R. Wang,et al., “Tumor organoid model of colorectal cancer (Review),” 26(2), 328
(2023).
14.
A. Habib, R. Mamistvalov, M. Malcov,et al., “Predicting colorectal cancer risk in FAP patients using patient-specific
organoids,” 32, 997–1007 (2025).
15. 1(2), 783–807 (2011).
16.
V. Backman, M. B. Wallace, L. T. Perelman,et al., “Detection of preinvasive cancer cells,” 406, 35–36 (2000).
17.M. Kassinopoulos, E. Bousi, C. Pitris,et al., “Correlation of the derivative as a robust estimator of scatterer size in
optical coherence tomography (OCT),” 8(3), 1598 (2017).
18.
H. B. Price, G. Song, W. Wang,et al., “Development of next generation low-cost OCT towards improved point-of-care
retinal imaging,” 16(2), 748–759 (2025).
19.
W.-C. Kuo, C.-M. Lai, and C.-Y. Chang, “Balanced detection for spectral domain optical coherence tomography,”
Opt. Express21(16), 19280–19291 (2013).
20.
I. Rubinoff, D. A. Miller, R. Kuranov,et al., “High-Speed Balanced-Detection Visible-Light Optical Coherence
Tomography in the Human Retina Using Subpixel Spectrometer Calibration,”
41, 1724–1734 (2022).
21.
D. A. Miller, M. Grannonico, M. Liu,et al., “Visible-Light Optical Coherence Tomography Fibergraphy of the Tree
Shrew Retinal Ganglion Cell Axon Bundles,” 43, 2769–2777 (2024).
22..
23.
H. Miyoshi and T. S. Stappenbeck, “In vitro expansion and genetic modification of gastrointestinal stem cells in
spheroid culture,” 8, 2471–2482 (2013).
24.
S. Yoshimoto, M. Taguchi, S. Sumi,et al., “Establishment of a novel protocol for formalin-fixed paraffin-embedded
organoids and spheroids,” 12(5), bio059882 (2023).
25.
D. M. Lyra-Leite, H. Fonoudi, M. Gharib,et al., “An updated protocol for the cost-effective and weekend-free culture
of human induced pluripotent stem cells,” 2, 100213 (2021).
26.
X. Lian, C. Hsiao, G. Wilson,et al., “Robust cardiomyocyte differentiation from human pluripotent stem cells via
temporal modulation of canonical Wnt signaling,” 109, E1848–1857 (2012).
27.
P. Hofbauer, S. M. Jahnel, N. Papai,et al., “Cardioids reveal self-organizing principles of human cardiogenesis,”
184, 3299–3317.e22 (2021).
28.
C.-W. Hsu, J. Cerda III, J. M. Kirk,et al., “EZ Clear for simple, rapid, and robust mouse whole organ clearing,”
11, e77419 (2022).
29. Wheater’s Functional Histology: A Text and Colour Atlas(Elsevier Health Sciences, 2006).
30.
J. Pyhtila, R. Graf, and A. Wax, “Determining nuclear morphology using an improved angle-resolved low coherence
interferometry system,” 11, 3473–3484 (2003).
31.
A. Wax, C. Yang, and J. A. Izatt, “Fourier-domain low-coherence interferometry for light-scattering spectroscopy,”
Opt. Lett.28, 1230–1232 (2003).
32.
R. N. Graf and A. Wax, “Nuclear morphology measurements using Fourier domain low coherence interferometry,”
Opt. Express13, 4693–4698 (2005).
33.
C. Photiou, A. D. Thrapp, G. J. Tearney,et al., “Classification of colon polyps with malignant potential using
statistical analysis of features extracted from ex vivo optical coherence tomography images,”
OnLine24, 108 (2025).