Source-linked AI summary
Eigenspectra optoacoustic tomography achieves quantitative blood oxygenation imaging deep in tissues
Stratis Tzoumas, Antonio Nunes, Ivan Olefir, Stefan Stangl, Panagiotis Symvoulidis, Sarah Glasl, Christine Bayer, Gabriele Multhoff, Vasilis Ntziachristos
TL;DR
Spectral corruption limits quantitative blood oxygenation imaging deep in tissue because wavelength-dependent fluence variations are difficult to predict. The paper introduces eMSOT, which models fluence using a few spectral base functions, and reports improved sO2 accuracy plus spatial agreement with perfusion and hypoxia maps.
Problem
Spectral corruption has limited accurate quantitative blood oxygen saturation imaging deep inside tissues because recovering tissue optical properties is complex and ill-posed.
Method
eMSOT models tissue light fluence as an affine combination of a few reference Eigenspectra and formulates sO2 estimation as constrained nonlinear spectral unmixing.
Results
eMSOT provided comparable to substantially better sO2 estimation accuracy than linear unmixing across more than 2000 simulations, phantoms, and animal measurements, with spatial agreement between tumor sO2 and histological perfusion and hypoxia maps.
Takeaways & Limitations
eMSOT quantitatively and noninvasively resolved blood oxygenation in phantoms, muscle, and tumors, including spatial patterns corresponding to perfusion-related hypoxia.
Takeaways & Limitations
eMSOT performs optimally on well-reconstructed portions of optoacoustic images and may face challenges in exact co-registration between in-vivo tumor images and ex-vivo histology.
Abstract
from arXiv · showhide
Light propagating in tissue attains a spectrum that varies with location due to wavelength-dependent fluence attenuation by tissue optical properties, an effect that causes spectral corruption. Predictions of the spectral variations of light fluence in tissue are challenging since the spatial distribution of optical properties in tissue cannot be resolved in high resolution or with high accuracy by current methods. Spectral corruption has fundamentally limited the quantification accuracy of optical and optoacoustic methods and impeded the long sought-after goal of imaging blood oxygen saturation (sO2) deep in tissues; a critical but still unattainable target for the assessment of oxygenation in physiological processes and disease. We discover a new principle underlying light fluence in tissues, which describes the wavelength dependence of light fluence as an affine function of a few reference base spectra, independently of the specific distribution of tissue optical properties. This finding enables the introduction of a previously undocumented concept termed eigenspectra Multispectral Optoacoustic Tomography (eMSOT) that can effectively account for wavelength dependent light attenuation without explicit knowledge of the tissue optical properties. We validate eMSOT in more than 2000 simulations and with phantom and animal measurements. We find that eMSOT can quantitatively image tissue sO2 reaching in many occasions a better than 10-fold improved accuracy over conventional spectral optoacoustic methods. Then, we show that eMSOT can spatially resolve sO2 in muscle and tumor; revealing so far unattainable tissue physiology patterns. Last, we related eMSOT readings to cancer hypoxia and found congruence between eMSOT tumor sO2 images and tissue perfusion and hypoxia maps obtained by correlative histological analysis.
RESULTS
eMSOT models tissue light fluence with three Eigenspectra and uses the resulting parameters for nonlinear sO2 estimation without explicit tissue optical-property maps. Simulations, phantoms, and animal experiments showed improved accuracy and spatially resolved oxygenation patterns in muscle and tumors.
- Eigenspectra model: Three Eigenspectra modeled simulated fluence spectra from arbitrary tissue structures with residuals typically below 1%.The second Eigenspectrum primarily tracked depth, while the first also reflected surrounding-tissue oxygenation.
- eMSOT inversion: eMSOT formulated blood sO2 estimation as nonlinear spectral unmixing using five or more wavelengths to recover hemoglobin concentrations and three Eigenfluence parameters.The fluence model combines the mean spectrum with three Eigenspectra weighted by spatially varying parameters.
- Validation: Across >2000 tissue simulations, eMSOT typically improved sO2 estimation accuracy 3–8 fold over linear unmixing at depths greater than 5 mm.eMSOT also produced lower sO2 estimation error with depth than the linear fitting method.
- Validation: In controlled mouse experiments, mean linear-unmixing error was 16–35%, versus 1–4% for eMSOT.These results indicated an order-of-magnitude improvement in mean sO2 accuracy.
- Muscle and tumor applications: eMSOT resolved muscle oxygenation gradients, measuring 60–70% near large vessels and 35–50% in soft tissue under normal breathing.Values changed with oxygen-breathing conditions and matched expected physiological states; ROI-2 values were interpreted as more related to capillary sO2.
- Muscle and tumor applications: In 4T1 tumors, eMSOT sO2 patterns corresponded spatially with perfusion and Pimonidazole-indicated hypoxia, while tumor cores showed limited response to oxygen challenge.The tumor measurements were related to histological Hoechst perfusion and Pimonidazole hypoxia assessments.
DISCUSSION
The discussion presents eMSOT as a noninvasive approach for quantitative deep-tissue blood oxygenation imaging, while identifying validation and scope boundaries. It reports improved accuracy over conventional spectral optoacoustic methods and spatial agreement with histological hypoxia patterns.
- eMSOT reformulates sO2 quantification in the spectral domain without estimating tissue optical properties.
- eMSOT produced quantitative, noninvasive, high-resolution blood oxygenation images in phantoms, muscle, and tumors.
- Tumor sO2 patterns showed close spatial correspondence with perfusion and hypoxia patterns measured by histological analysis.
- More than 2000 tissue simulations showed eMSOT provided comparable to substantially better sO2 estimation accuracy than linear unmixing.
- In-vivo validation was challenging because available comparison methods are invasive, indirect, tracer-dependent, or lack spatial information.
- The method was optimized for well-reconstructed image regions and quantifies blood sO2, but not absolute blood volume.
ACKLOWLEDGEMENTS
The authors acknowledge funding from European and German research programs and thank colleagues who assisted with phantom preparation and scientific discussions.
- Vasilis Ntziachristos acknowledges support from an ERC Advanced Investigator Award and the European Union FAMOS project.
- Stratis Tzoumas was supported by the DFG GRK 1371 grant.
- The authors thank Elena Nasonova, Karin Radrich, Amir Rosenthal, and Juan Aguirre for assistance and discussions.
METHODS
The methods combine multispectral optoacoustic measurements with an Eigenspectra light-fluence model and constrained nonlinear inversion. The study uses simulations and animal imaging to estimate blood sO2 while accounting for wavelength-dependent fluence.
- Animal preparation and handling: Animal experiments used orthotopic 4T1 tumors in eight female athymic Nude-Foxn1 mice imaged in vivo with MSOT.
- Animal preparation and handling: The animal study used no randomization, blinding, or statistical methods.
- MSOT: MSOT acquired cross-sectional images with a 256-element transducer array at 10 Hz using 21 wavelengths from 700 to 900 nm.
- eMSOT method and sO2 maps: Measured optoacoustic signals are related to oxy- and deoxyhemoglobin concentrations through wavelength-dependent extinction coefficients and light fluence.
- eMSOT method and sO2 maps: The normalized fluence spectrum is defined by dividing wavelength-dependent fluence by its across-wavelength Euclidean norm.
- eMSOT method and sO2 maps: Space-only scaling factors do not affect sO2 because the estimate uses relative oxy- and deoxyhemoglobin concentrations.
- The Eigenspectra Model for Light Fluence: eMSOT models tissue fluence as an affine combination of a mean spectrum and a few base spectra derived using PCA.
- The Eigenspectra Model for Light Fluence: The three-Eigenspectra model was tested on heterogeneous media with varying optical properties and oxygenation values.
SUPPLEMENTAL MATERIAL
The Eigenspectra model was validated across simulated and experimental tissue conditions, showing that a compact affine model can capture wavelength-dependent light-fluence spectra. eMSOT generally improved sO2 estimation over linear unmixing, including in deep tissue and blood phantoms.
- Forward-model validation: The Eigenspectra forward model produced small errors across random, low-variation, vessel-like, melanin-containing, and wavelength-dependent scattering tissue simulations.The results support using three Eigenspectra to capture spectral variability largely independently of optical-property distributions.
- Forward-model validation: Monte Carlo simulations in the ballistic regime produced a forward-model fitting residual of 0.61±0.22%.The test used four multilayered tissue layers with different oxygenation levels and optical properties.
- Experimental validation: In-vivo and post-mortem mouse measurements showed low fitting residuals between measured fluence spectra and the Eigenspectra model.A reference absorber in a capillary tube enabled experimental fluence measurements under different physiological conditions.
- Numerical validation: More than 2000 simulations showed mean sO2 estimation errors of 2.4%–3.4% for mean tissue oxygenation between 30% and 80%.Approximately 97% of cases had sO2 error below 10%; errors were larger below 30% mean oxygenation.
- Comparison with linear unmixing: 88% of simulated datasets had lower mean estimation error with eMSOT than with conventional linear unmixing.For tissue depths greater than 5 mm, eMSOT typically provided 3- to 8-fold enhanced sO2 estimation accuracy.
- Phantom validation: Blood-phantom experiments showed eMSOT errors typically below 10%, whereas linear unmixing errors could reach 30%.The comparison covered eight phantoms with varying background and target oxygenations.
Supplementary Note 5: Application of eMSOT on experimental tissue images
Experimental tissue application estimates Eigenfluence parameters from high-quality image regions using constrained inversion and interpolation. The corrected spectra then support sO2 estimation where raw-spectrum linear fitting is inaccurate.
- Prior estimation: Prior Eigenfluence estimates are computed from a 3D light-propagation model and 20 neighboring MSOT slices surrounding the analyzed central slice.This provides robust priors when sO2 varies substantially within the illuminated three-dimensional volume.
- Grid selection: eMSOT requires a high-intensity, high-SNR image region without visible reconstruction artefacts for grid placement.The selected grid spectra are used by the constrained inversion algorithm to estimate m1(r), m2(r), and m3(r).
- Map reconstruction: After estimating the Eigenfluence parameters at grid points, eMSOT computes intermediate Eigenfluence maps by cubic interpolation.The maps can be compared across post-mortem, 20% O2, and 100% O2 breathing conditions.
- Spectral correction: eMSOT decomposes raw optoacoustic spectra into corrected normalized absorption spectra and estimated light-fluence spectra.Linear fitting on raw spectra produces high residuals and inaccurate sO2 estimates in the illustrated tissue data.
Supplementary Note 6: Imaging tumor hypoxia with eMSOT and histological validation
Tumor imaging combined eMSOT sO2 maps with anatomical and histological measurements to examine perfusion-related hypoxia. The analyses resolved intratumoral oxygenation differences and hypoxia-related heterogeneity, while distinguishing perfusion from diffusion hypoxia.
- Tumor imaging: Eight mice bearing orthotopic 4T1 mammary tumors were imaged with MSOT at transverse slices through the lower abdominal tumor region.Tumor regions were identified anatomically and co-registered with subsequent cryosliced histology.
- Perfusion and oxygenation: Two tumors showed different eMSOT sO2 levels alongside distinct core-to-boundary Hoechst perfusion intensity ratios of 48% and 19%.The lower-perfusion tumor core corresponded to lower eMSOT sO2 values.
- Histological validation: eMSOT tumor sO2 patterns closely corresponded to hypoxic areas identified by Pimonidazole staining and perfusion patterns identified by Hoechst 33342.The method distinguished high and low hypoxia levels and revealed intratumoral spatial heterogeneity.
- Scope of hypoxia detection: The histological comparison supports eMSOT detection of perfusion-related hypoxia within solid tumors.Pimonidazole-positive cellular hypoxia can also arise from diffusion hypoxia, which does not produce an eMSOT signal.
- Limitations: Exact co-registration between in-vivo eMSOT tumor images and ex-vivo histology may be technically challenging.This limits direct spatial correspondence between the two measurement modalities.