Light sheet luminescence imaging with Cherenkov excitation in thick scattering media

Abstract

Light scattering leads to a severe loss of axial and transverse resolution with depth into tissue, limiting accuracy and value of biomedical luminescence imaging techniques. High-resolution imaging beyond a few-millimeter depth is prohibited because diffusive transport dominates beyond a few scattering distances. In this study, light sheet imaging through scattering media is demonstrated using a radiotherapy linear accelerator to deliver well-defined thin scanned sheets of X-rays. These sheets produce Cherenkov light within the medium, which in turn excites luminescence of an optical probe across the sheet plane. This luminescence can then be imaged by an intensified camera positioned perpendicularly to the sheet plane. The precise knowledge of the light sheet position within the medium allowed efficient correction for attenuation of the signal with depth as well as spatial deconvolution of the excitation light. Together these methods allowed for the first time a high-resolution imaging of tissue-equivalent phantoms up to 3 cm thick, yielding the precise position and shape of luminescent lesions located deep in tissue without the need for non-linear image reconstruction.

Deep-tissue fluorescence molecular imaging with optical transillumination or epi-illumination suffers from poor spatial resolution because of the tissue’s severe fluorescence photon scattering. The distribution of excitation light through a scattering sample is defined by multiple scattering events, which result in a diffuse light transport process. Consequently, current deep tissue optical imaging approaches lack accurate information about the light fluence at the locations of luminescent probes within the tissue, and that inherently limits their localization precision. The accuracy of diffuse-tomographic methods [1] can be improved by integrating prior structural information [2], but the inverse problem still remains highly ill-posed in diffuse optical imaging. While different mathematical techniques have been demonstrated to improve the imaging resolution [3–5], their performance is still inherently limited by the light scattering and compounded by specimen complexity and topology. In this study, we aimed to shift the paradigm of unknown source location by using a shaped radiotherapy beam to induce Cherenkov emission from within the tissue itself, as illustrated in Fig. 1(a). The Cherenkov radiation becomes a localized excitation light source, since the mega-Volt X-ray beams do not suffer from significant intensity loss while propagating through the tissue. Particularly in this study, we show that with prior knowledge of where the excitation light sheet was located within the scattering object, a nearly linear correction can be applied to both the attenuation and spatial spread of the emission excited by that light sheet. Our approach is directly analogous to the light sheet microscopy used in thin specimens, which is known for its superior resolution combined with advantages of fast imaging in wide field [6, 7].

Fig. 1.

(a) The Cherenkov light sheet luminescence imaging principle with frontal delivery and dorsal imaging of emission; (b) the triggering sequence of the intensified camera with a 3.7 μs pulse followed by a 20–45 μs phosphorescence decay.

The polychromatic Cherenkov emission inside an object is directly proportional to the deposited radiation dose [9]. In mega-Volt (MV) X-radiation therapy delivery, the depth dose curve rises within the superficial build-up region to the maximum in 1.5–3 cm depth, and then decays exponentially with depth. However, with a narrow flat X-ray beam the buildup distance shortens to several mm. Furthermore, the spatial extent of the dose buildup is comparable to the optical diffusion attenuation. Thus, the optical fluence of Cherenkov radiation produced by an X-ray sheet is nearly constant over a few cm, along the X-ray beam path after the first 3 millimeters (Fig. 2(d)). By shaping the X-ray beam into a thin sheet and observing the Cherenkov-excited luminescence orthogonally to the sheet beam, we were able to capture a luminescence image emanating from only a single plane within the tissue. Analogous to light sheet microscopy, a series of luminescence images were taken for varied axial positions (depths) of the Cherenkov light sheet in the sample (Fig. 2(b)). The resulting three-dimensional image stack was then subject to the 3D luminescence distribution reconstruction. The latter step consists of the deconvolution and depth-variant attenuation correction, both exploiting the knowledge of the excitation plane position. This procedure is dramatically different from diffuse illumination and detection (Fig. 2(a)) where the incident light is attenuated diffusely, resulting in significant dynamic range differences and uncertainty about the excitation fluence with depth.

Fig. 2.

Schematic comparison of (a) standard diffuse light excitation method and (b) frontal beam Cherenkov-excited luminescence scanning imaging (CELSI); the Cherenkov intensity profiles (red curves) measured orthogonally (c) and in parallel (d) to the X-ray sheet beam. Inset (c) includes normalized transverse profile of deposited X-ray dose (blue), calculated from experimental data using Eq. 5. In (d) the intensity profile is dictated by the x-ray build up, reaching maximum 9 mm into the phantom from the right. The inset of (d) depicts the Cherenkov emission spectrum (red) and a normalized excitation/emission spectrum of PtG4 probe (green).

Based on our previous proof-of-principle reports on Cherenkov-excited luminescence scanned imaging (CELSI) [9], here we present the frontal sheet beam scanning methodology that is more suitable for imaging with wider field of view. Our optimized scheme requires the use of only a single sheet-beam scan and a 2D imaging detector, instead of using three orthogonal scans and a point detector [9]. This new geometry renders the frontal CELSI more suitable for high-throughput scanning of multiple animals and/or with multiple fluorescent lesions. The aim of this study was to demonstrate the capabilities of deep-tissue frontal beam CELSI in terms of depth sensitivity, spatial resolution, and multiple objects recovery using a tissue-equivalent phantom. The complete 3D image reconstruction procedure that we employed makes use of linear depth-dependent attenuation correction and spatial deconvolution, both of which are possible only with precise knowledge of the location of the excitation sheet - the unique capability intrinsic to CELSI.

A phantom was created to simulate multiple luminescent probe-marked lesions located several centimeters deep in tissue. Three 400 μL tubes (diameter 6 mm) containing 25 μM solution of oxygen sensitive near-infrared (NIR) phosphorescent probe PtG4 [9–11] in the standard phosphate buffer (PBS, pH 7.2) were placed at varying depths (z = −10 mm, −20 mm and −30 mm, resp.) in a rectangular container (110 × 60 × 65 mm). The container was filled to z = 0mm level with 1% v/v Intralipid© (Fresenius Kabi, Uppsala, Sweden) PBS, solution as a tissue-mimicking scattering medium (μ_a ≈ 0.0026 mm⁻¹, μ_s’ ≈ 1.0 mm⁻¹ at 800 nm) [12].

The frontal beam CELSI scanning of the phantom was performed using a clinical radiotherapy accelerator (Varian Linac 2100CD, Varian Medical System, USA). The accelerator was set to generate a pulsed beam with the following parameters: photon energy 6 MV; 9000 monitor units per scan (MU·scan⁻¹); repetition rate 360 Hz; 3.25 μs pulse duration. Multi-leaf collimators (MLCs) were used to shape and vertically translate the beam profile (200 mm width × 5 mm height, 400 steps per scan, 0.2 mm·step⁻¹). This delivery plan is approximately equivalent to an average dose of 2.7 Gy. The MLC scan was performed by delivering a custom dynamic dose plan, created in MLC Code [13].

The luminescence was imaged by a gated, intensified charge-coupled device (ICCD, PI-MAX4 1024i, Princeton Instruments, USA). A 45° front-surface mirror was placed above the top surface of the sample, allowing horizontal mounting of the camera with 135 mm f/2 lens in the distance of 2 m from sample. Time-gated acquisition of the luminescence signal (50 μs integration time, 3.7 μs delay after each X-ray pulse; Fig. 1(b)) provided an effective means to eliminate the detection of background Cherenkov radiation and auto-fluorescence. The image intensifier was triggered using the current output signal from the linear accelerator target. The images were acquired with 100× gain on the ICCD intensifier, 1062 photon accumulations on chip yielding 300 frames/scan, and 4×4 pixel hardware binning upon readout. The CELSI image stack was corrected by background subtraction and smoothed using median filtering with a 2×2×2 pixel neighborhood.

In the image, the steady state luminescence intensity distribution, $I$, can be expressed as a convolution of the original luminescence distribution, $L$, with a radiation transport kernel, $\Phi$, which accounts for both the photon diffusion and source properties. The inverse problem needs to be solved by non-linear iterative methods when the light source is external to the tissue because of the highly ill-posed nature of the inversion [14]. However, in the case of CELSI, the known position and shape of the excitation light allowed us to factorize the radiative transport kernel into a series of sub-kernels with lower dimensionality. In turn, the inverse problem could have been solved by a simple stepwise deconvolution of the acquired dataset using the aforementioned sub-kernels. These sub-kernels were defined by experimentally derived effective attenuation and scattering coefficients assumed equal for all the wavelengths in the luminescence spectrum. In our experiments, the luminescent objects were surrounded by a homogeneous, heavily scattering medium, making the Cherenkov emission nearly uniform, and allowing for high oversampling in image acquisition (voxel size ≪ desired vertical resolution). Our phantom study was designed to fulfill these requirements, thereby closely matching the real tissue-imaging situation. These experimental characteristics allowed us to factorize the transform kernel, also known as propagator, $\mathrm{\Phi }$ into three parts:

1. the source component responsible for the vertical (Z-axis) blur of the luminescence intensity profile; this component can be expressed as a finite concentration of the Cherenkov radiant power ${\mathrm{\Phi }}_X$ along the transversely symmetric Gaussian-shaped X-ray beam of width $b$, propagating in the medium with an effective attenuation coefficient ${\mu }_{\mathit{eff}}$ at depth $z$:

   $${\mathrm{\Phi }}_X\left(z\right)=\exp \left(-\frac{z^2}{2b^2}\right)\mathrm{*}\exp \left(-\left|{\mu }_{\mathit{eff}}z\right|\right);$$ (1)
2. the component describing the attenuation of the luminescence intensity attenuated due to the losses caused by the emission photon diffusion as it escapes up through distance, $\mathrm{z}$:

   $$T\left(z\right)=\exp \left(-{\mu }_{eff}z\right);$$ (2)
3. and the component related to the loss of the image contrast due to the luminescence diffusion through scattering media from depth $z$, expressed as a convolution of the true luminescence intensity profile with Gaussian blurring:

   $${\mathrm{\Phi }}_M\left(x,y,z\right)=\frac{1}{\sigma \left(z\right)\sqrt{2\pi }}\exp \left(-\frac{(x+y{)}^2}{2\sigma (z{)}^2}\right),$$ (3)

where $\sigma (\mathrm{z})$ is the parameter quantifying the loss of the frontal Gaussian point spread function (PSF) fidelity as a function of depth. The forward problem can therefore be described as:

$$I=\left(L\mathrm{*}{\mathrm{\Phi }}_M\right)T\mathrm{*}{\mathrm{\Phi }}_X$$ (4)

We implemented the stepwise deconvolution algorithm using the Mathematica environment (Wolfram, USA). As outlined in Fig. 3., we attempted to invert and solve the problem (Eq. 4) by eliminating the photon transport components (Eq. 1 - Eq. 3.) using iterative Lucy-Richardson method. Starting at the rightmost side of Eq. 4, we first deconvolved each column of the acquired dataset with the source kernel ${\mathrm{\Phi }}_X$ (Eq. 1). In the second step, we corrected the deconvolved distribution for the depth attenuation. This was achieved through a numerical division of each column in the dataset by the intensity component as defined in Eq. 2. In the last step, each frontal slice of the corrected data stack was deconvolved with a depth-variant Gaussian kernel (Eq. 3).

Fig. 3.

(a) The diagram of CELSI post-processing sequence; (b)-(d) sequence of mid-saggital section of the phantom displayed in different post-processing steps.

The solution of the inverse problem required an input of optical parameter ${\mu }_{\mathit{eff}}$, source parameter $b$, and depth-dependent frontal PSF width $\sigma (\mathrm{z})$. In order to extract these parameters, we performed initial X-ray beam scan and recorded the intensity of the Cherenkov emission from a scattering phantom prior to the insertion of tubes containing the phosphorescent probe. The ICCD intensifier gating was set to follow the LINAC trigger signal with a delay of 3.7 ns to capture most of the Cherenkov light, integrating the signal on CCD for 4 μs. As the beam passed through the object’s upper surface boundary, the Cherenkov signal first rose in a Gaussian manner and then decreased exponentially. We then extracted the parameters ${\mu }_{\mathit{eff}}=0.09{\mathrm{m}\mathrm{m}}^{-1}$ and $b=2.8\mathrm{m}\mathrm{m}$ by fitting the resulting curve with the transverse propagation function ${\mathrm{\Phi }}_{CH}(z)$ of the source

$${\mathrm{\Phi }}_{CH}(z)=\mathrm{e}\mathrm{x}\mathrm{p}\left(-\frac{z^2}{2b^2}\right)\mathrm{*}H(z)\mathrm{e}\mathrm{x}\mathrm{p}\left(-2{\mu }_{eff}z\right),$$ (5)

where $H(z)$ is a Heaviside step function. Finally, the depth-dependent PSF parameter $\sigma (\mathrm{z})$ was calculated from the proportionality of the Gaussian amplitude and width:

$$\sigma \left(z\right)=\frac{1}{\sqrt{2\pi }\exp \left(-{\mu }_{eff}z\right)}.$$ (6)

The stepwise deconvolution of photon transport kernels from Eq. 4 resulted in a three-dimensional image of the phantom (Fig. 4.), displaying all three luminescent objects at their correct depths (z₁ = −10.2 mm, z₂ = −20.4 mm z₃ = −30.6 mm), and with correct relative intensities (I₁ = 1, I₂ = 0.727, I₃ = 0.737). The diameter of the recovered phosphorescence signal measured longitudinally at full-width half maximum intensity was 12 mm at 10 mm depth and rose to 17.7 mm in 30 mm depth. The depth detection limit was governed by the fast decrease in the signal-to-noise ratio (SNR) with depth. The estimated SNR of the phosphorescence peaks in our measurement was equal to 52, 13, and 4. At depths > 40 mm the detected noise signal with small non-zero mean became to be heavily amplified in the reconstruction process. Without substantially increasing the brightness of the luminescent phantoms, this depth appeared to be the imaging limit for CELSI in this particular experiment.

Fig. 4.

The recovered three-dimensional image of luminescence intensity inside a tissue-equivalent phantom.

Lastly, we compared the image quality of the CELSI with the phosphorescent probe to that using the same phantom and commercial imager (IVIS 200 Spectrum CT, PerkinElmer, USA). The image of the phosphorescence acquired form the top surface (excitation wavelength 640 nm, emission wavelength 780 nm, integration time 2 s) is shown in Fig. 5.(a) and compared with the vertical maximal projection image obtained using CELSI 3D reconstruction. The IVIS image exhibits a marked drop in the signal intensity (measured at the centers of the three objects): $\mathrm{e}\mathrm{x}\mathrm{p}\left[-{\mu }_{eff}\left({\lambda }_X\right)d-{\mu }_{eff}\left({\lambda }_M\right)d\right]$, accounting for both excitation and emission attenuation coefficients $-{\mu }_{eff}\left({\lambda }_X\right)$ and $-{\mu }_{eff}\left({\lambda }_M\right)$, resp. (Fig. 5(d)). On the other hand, the projected raw CELSI image (Fig. 5.(b)) shows the intensity drop due to the luminescence emission attenuation only. Most importantly, the post-processed image in Fig. 5.(c) shows all the three luminescent objects with approximately equal intensity and shape, which corresponds well to the true distribution of the phosphorescence.

Fig. 5.

(a) top-view luminescence image obtained by IVIS Spectrum CT system; maximum intensity-projected top view of the (b) raw and (c) post-processed CELSI image. (d) Vertical projection of maximum intensity values of raw and post-processed CELSI image at mid-sagittal plane (red), compared to the mid-sagittal intensity profile as detected by IVIS Spectrum CT system (black). Each profile in (d) was normalized to the maximum intensity.

The main advantage of the frontal light sheet CELSI technique is the capability to perform 3D molecular imaging through several centimeters of tissue while retaining high-spatial resolution. The purpose of this work was to demonstrate such tomographic imaging of multiple luminescent objects in tissue-mimicking scattering media at depths reaching up to 30 mm. We developed an iterative image reconstruction scheme with correction for the image depth-dependent spatial blur and signal attenuation. The scheme showed excellent performance in image recovery thanks to the depth attenuation correction. Further analysis of SNR and contrast recovered must also be coupled to dose delivered, and reductions in dose are easily possible with proportionate reductions in SNR, but likely still viable [9].

In this paper, we presented an illustrative image recovery principle that is valid for homogenous phantoms with simple geometry. To allow an in-vivo fluorescence imaging of objects with an arbitrary shape, the reconstruction process can be performed by means of finite-element analysis method using such as NIRFAST [14]. However, it is important to recognize that a comparison to diffuse tomographic recovery is challenging given the fact that most quantitative papers have their own systems and algorithms that cannot be utilized by others, and also the fact that the SpectrumCT tomographic capability has never been systematically evaluated in the peer reviewed literature. Sub-surface tomography is well known to be limited to about 1.5cm of tissue depth, and quantitative recovery is non-linear with depth into tissue [15].

It is also possible to use additional information about the object’s optical properties and topology (e.g. computed tomography data) instead of assuming the optical homogeneity of the object. Once tested with a larger selection of the useable luminescent probes [16], we believe that the CELSI technique can offer a viable modality for small-animal imaging and possibly for a functional imaging during clinical radiotherapy.