Open Access
Add to BibSonomy
Abstract
Ultrasound (US) is a valuable tool for imaging soft tissue and visualizing tumor contours. Taking the benefits of US, we presented an integrated dual-modality imaging system in this paper that achieves three-dimensional (3D) bioluminescence tomography (BLT) with multi-view bioluminescence images and 3D US imaging. The purpose of this system is to perform non-invasive, long-term monitoring of tumor growth in 3D images. US images can enhance the accuracy of the 3D BLT reconstruction and the bioluminescence dose within an object. Furthermore, an integrated co-registered scanning geometry was used to capture the fused BLT and US images. We validated the system with an in vivo experiment involving tumor-bearing mice. The results demonstrated the feasibility of reconstructing 3D BLT images in the tumor region using 3D US images. We used the dice coefficient and locational error to evaluate the similarity between the reconstructed source region and the actual source region. The dice coefficient was 88.5%, and the locational error was 0.4 mm when comparing the BLT and 3D US images. The hybrid BLT/US system could provide significant benefits for reconstructing the source of tumor location and conducting quantitative analysis of tumor size.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
The use of small animal imaging in oncology research has been expanding, with various imaging modalities like magnetic resonance imaging (MRI), positron emission tomography, and optical imaging (including bioluminescence tomography imaging and fluorescence imaging) being regularly utilized [1–4]. Optical techniques are popular in preclinical and basic research due to their affordability, capacity to handle large sample sizes, and high sensitivity in molecular-level assessments. However, each modality has its limitations, making multimodal imaging increasingly essential to combine the advantages of different techniques. As such, optical techniques incorporating prior structural information have been successfully developed. As a result, optical techniques incorporating prior structural information have been successfully developed. Recent preclinical studies have demonstrated the benefits of dual imaging techniques, such as combined MRI/bioluminescence imaging [5], computed tomography (CT)/fluorescence imaging [6], and CT/bioluminescence tomography (BLT) imaging [7,8]. Our study explored the combined use of BLT and ultrasound (US) imaging in a subcutaneous tumor model. Incorporating structural information from US imaging is particularly beneficial in creating more accurate heterogeneous models, as US provides more detailed information in soft tissues than other anatomical imaging techniques like CT or MRI.
BLT is a valuable noninvasive imaging method used to assess treatment effects as it can provide quantitative information about the location and timing of tumor growth in live mice [9]. Over the past decade, BLT systems have been widely used to monitor and study various biological processes, such as detecting cancer cells or quantifying drug distribution in pre-clinical studies [10–12]. The BLT approach relies on the genetic modification of target cells and molecules within an animal model, leading to the cellular expression of a luciferase enzyme similar to that found in fireflies [13]. Upon injection of the appropriate substrate and the presence of ATP and oxygen, visible/NIR photons are emitted from the target location, and part of these photons can penetrate through the surrounding tissue, making them available for external detection by a highly sensitive camera. Thus, bioluminescence molecular imaging systems can non-invasively capture the distribution images of bioluminescent photons. We can gather more quantitative data by integrating bioluminescence imaging with other imaging techniques like CT, MRI, or US that provide complementary information, such as tumor volume. This data can be used to create more precise treatment planning models and measure the effectiveness of the therapy [14].
If BLT is merged with 3D anatomical structure information gathered from CT or MRI scans, it becomes possible to determine the location of the internal bioluminescence light source, which aligns with the lesions, using tomographic algorithms. This can be achieved by non-invasively detecting bioluminescence images on the surface, which allows for the identification of a bioluminescent source within a small animal [15]. The challenge for BLT is to accurately reconstruct 3D volumetric maps of bioluminescent source distributions within living animals from data collected from the animal's external surface [16]. The BLT reconstruction is a generally ill-posed and non-unique problem, which usually requires additional information to achieve acceptable reconstruction results [17]. Studies have shown that prior structural information obtained from MRI or CT scans can benefit BLT or fluorescence tomography reconstruction when used as a structured regularization [18,19]. The precise structural data obtained through MRI or CT scans can be used with these systems, enabling the segmentation and assigning of published optical properties to specific organs and other parts of the animal's volume [20]. However, the operational cost is high, acquisition times are lengthy, access is limited due to high user demand, and safety considerations, including ionizing radiation, must be taken into account when using these modalities [21,22]. Therefore, this study used US images as prior structural information and improved the accuracy of tumor segmentation in medical images. To quantitatively evaluate the quality of the reconstruction results with anatomical information like US imaging, the location error (LE), and dice coefficient (DICE) was employed in this study. A high DICE indicates that the shapes and sizes of targets are well reconstructed. LE shows the difference between the reconstructed source center and the actual source center. For these reasons, using 3D US images as prior structural information allows for faster segmentation and better overall performance than other methods [23,24].
Using US images as prior information has several benefits. US imaging is non-ionizing, provides real-time imaging, and is relatively inexpensive [25–27]. Recently, US imaging has gained increasing attention in optical imaging. Many researchers have used US images as prior information to guide and constrain the BLT reconstruction [23,28–30]. Bal and Schotland used US imaging as the anatomical information to solve an inverse problem of the diffusion equation to obtain BLT images. The experimental results showed that the US prior information performed well and could accurately locate results [28]. Czernuszewicz, T. J. et al. developed a new platform that uses high-frequency US for anatomical imaging of tissue and bioluminescence imaging for both anatomical and functional imaging of tumors [23]. Additionally, our team has completed preliminary research, which involved simulating 3D Fluorescence Diffuse Optical Tomography (FDOT) using a phantom and an ex vivo test [31]. The alignment was complicated because a handheld US transducer was used to acquire the US imaging. Due to the complexity of connecting two different modalities, we need to find an effective method to reduce the time spent on aligning functional and anatomical imaging.
This study developed a new 3D imaging system for small animals that accomplishes 3D BLT using multi-view bioluminescence images and 3D US imaging. To validate the design and assess the reproducibility of the imaging, we conducted studies on mice with tumors to evaluate changes in drug metabolism over time. We successfully measured and accurately located the distribution of a bioluminescent substance when it was associated with the US anatomy. In sum, we enhanced the accuracy of BLT reconstructions by using multiple-view bioluminescence images and introduced a customized mouse holder that enabled co-registered BLT and US imaging. The DICE was 88.5%, indicating a high level of agreement. Furthermore, the locational error, which measures the discrepancy in positioning, was less than 0.4 mm, confirming the accuracy of our imaging technique.
2. Materials and methods
2.1 Imaging system and instrumentation
The BLT/US dual-modality imaging system is composed of an optical and US subsystem. The optical system includes an electron-multiplied charge-coupled device (EMCCD) camera (ProEM 512 B-eXcelon, Princeton Instruments, Inc.) that is cooled to −70 °C, and a lens (Model: SIGMA 24 mm f/1.8 EX DG) with a focal length of 24 mm and an aperture that ranges from f/1.8 to f/22. The camera captures images and stores them on a computer. The US system includes a custom single-element US transducer, a US pulser-receiver (US-Wave, Lecoeur Electronique, Chuelles, France), and an oscilloscope (12-bit ADC resolution, 125 MS/s sampling rate) (HDO4034, Teledyne LeCroy, Inc.). The US transducer has a center frequency of 7.8 MHz and a focal length of 15 mm. A 360-degree scanning module, which includes spinning scanning motors and a mouse holder, connects the two subsystems. The optical and US systems are connected to a computer via a USB connection, which allows the computer to receive and process images from both systems. A diagram of the dual-modality imaging system is shown in Fig. 1. The contour of the mouse with a tumor is provided by a self-made US scanning module that moves in the horizontal plane. The homemade animal holder keeps the mouse stable during the imaging process. A semi-cylindrical Polyvinylchloride (PVC) film holds the mouse's body, while grooves reduce the mouse's displacement. An infrared heating device is used to maintain the mouse's body temperature. The US images are obtained by scanning multiple sections, and the 3D profile of the mouse is constructed by integrating different sections.
Fig. 1. Dual-modality bioluminescence/ultrasound 3D 360-degree imaging system. (a) Schematic of the system, (b) photograph of a bioluminescence and US subsystem with a mouse holder in the middle, indicated by a yellow circle, and an anesthesia tube, indicated by a red arrow.
2.2 Tumor model
The Mouse Tongue Carcinoma, 4-nitroquinoline 1-oxide (4NQO) induced (MTCQ1) cell line, which was derived from an oral squamous cell carcinoma (OSCC) and obtained from the Japanese Collection of Research Biosources Cell Bank, was established from Mouse tongue carcinoma 4NQO [32]. This cell line was maintained in Gibco Roswell Park Memorial Institute (RPMI) medium (GIBCO Invitrogen Inc., Carlsbad, CA, USA) with 10% fetal bovine serum (HyClone Thermo, Waltham, MA, USA), 50 µg/mL of penicillin/streptomycin (Sigma-Aldrich Co., St. Louis, MO, USA), 2 mM of l-glutamine (Sigma-Aldrich Co., St. Louis, MO, USA) and incubated at 37 °C in a humidified atmosphere with 5% CO2. The cells were passaged every two days.
BALB/c female nude mice (5–6 weeks, 18–22 g) were purchased from the National Laboratory Animal Center (Taipei, Taiwan) and housed in a temperature-controlled environment (22 ± 1°C) with a 12-hour light/dark cycle and unlimited access to food and water. The Institutional Animal Care and Use Committee of the National Yang Ming Chiao Tung University (approval number: 1090306r) approved the MTCQ1 tumor establishment experiments. For these experiments, 1 × 106 MTCQ1 cells in a total volume of 100 µL were injected subcutaneously into the right hindlimb of BALB/c female nude mice.
2.3 BLT imaging
The mice were anesthetized with 2% isoflurane and injected intraperitoneally (IP) with 100 µL of D-Luciferin (Goldbio). The bioluminescence imaging images were captured up to 60 minutes after the luciferin injection to record the most intense signal using our custom-made BLT/US dual-modality imaging system. Eight bioluminescence surface images from different viewing angles (45-degree increments for a full 360-degree range) were acquired by rotating the position of the mice. Three bioluminescence surface images (0°, 45°, and -45°) were applied to the BLT reconstruction algorithm. Bioluminescence surface images from 90, 180, and 270-degree views of projections were eliminated due to the minimal signals. Moreover, kinetic curve was created by imaging the mice every 10 minutes up to 60 minutes after the luciferin injection. Initially, the exposure time was 1 minute, but as the tumors became increasingly brighter, the imaging parameters were adjusted to maximize the bioluminescence signal without overloading the detector.
2.4 Image co-registration
The dual imaging in this system incorporates US tomography and bioluminescence intensity projection images. The alignment of the EMCCD on the animal holder must be carefully managed to identify the rotational axis in the optical data. The animal is kept in a rigid pose, at a fixed position relative to the animal holder, during and between the imaging sessions. We acquired eight bioluminescence images (0° to 360° projections with a 45-degree increment). Following this, we mapped the bioluminescence images onto a 3D mesh surface generated from the US images. For the co-registration of the two images, we followed three steps: (1) convert both the US and bioluminescence images from pixel to mm scale. (2) From US tomography, we can obtain the contour of the object. Since the optical (BLT) rotation axis is known, we can map the optical rotation center in US tomography. (3) The bioluminescence data can be delivered in a coordinate centered on the rotation axis. (4) The bioluminescence projection images can be mapped onto a 3D mesh surface generated from the US images.
2.5 3D mesh construction
MATLAB was used to make a US and optical data processing in this work. The 3D BLT reconstruction was based on the structural prior information provided by 3D US images. We utilized acoustic gel as the acoustic media for US scanning, which resulted in clear and high-quality B-mode images with integrated scanning geometry. The time it takes to report a 3D US scan can range from 30 to 40 minutes, depending on the mice's size and the case's complexity. US B-mode images of 25 sections were obtained; the interval between each section was 1 mm, and the contour and organ segmentation were manually performed by outlining the organs in each slice of the 3D US image. The profiles of the body and tumor from 25 sections were positioned in a 3D Cartesian coordinate system to outline the mouse's contour. Figure 2(a) displays 3D models of a mouse and a tumor, created by layering multiple 2D US images from different sections. A 3D mesh was constructed using finite element analysis, with the maximum mesh thickness set at 0.1 mm and a total of approximately 17,000 nodes. The 3D BLT reconstructions were displayed in a 3D space using 3D Slicer software. 3D Slicer is an open-source medical software that can reconstruct and visualize various medical images. Figure 2(b) shows the flowchart of the 3D BLT. The stopping criterion for the algorithm is when the projected error change is less than 2% and DICE exceeds 70%, or when the iteration count reaches the limit set by the user.
Fig. 2. 3D BLT data analysis method. (a) Procedure for making the contour of the mouse, (b) a flowchart of US data processing, and the 3D BLT reconstruct algorithm.
2.6 3D BLT
We would set the initial optical properties on the mesh, which include scattering coefficient (µs), absorption coefficient (µa), and fluorescence quantum efficiency (η). For heterogeneous model, the optical properties of each organ are listed in Table 1. For homogenous model, the optical properties of the mice are assigned ${\mu _a}$ = 0.01 mm−1, $\mu _s^{\prime}$ = 1 mm−1. The fluence rate of the bioluminescence signal, originating from a source inside a tissue sample, was computed by modifying the open-source light transport model, Near Infrared Fluorescence and Spectral Tomography (NIRFAST). The mesh generation tool in NIRFAST was subsequently utilized to divide the mice body and tumor region into tetrahedral meshes. In this situation, the fluence rate of bioluminescent photons detected at the tissue sample's surface can be calculated by invoking the forward model in NIRFAST. This model solves the diffusion approximation of the radiative transport equation in the frequency domain. Equation (1) and (2) were used to calculate the light field and bioluminescence [33,34].
Table 1. Optical parameters of mouse organ
In Eq. (3), ${G_\mathrm{\lambda }}$ is the sensitivity matrix describing the changes of surface fluence rate ${\varphi _\mathrm{\lambda }}$ related to source S for a given wavelength λ, which can be constructed from prior knowledge of the optical properties of the imaged object, and ${\mathrm{\omega }_\mathrm{\lambda }}$ is the system-specific light source spectrum.
To recover the internal distribution of bioluminescence, we need to solve the linear relationship described by Eq. (3). However, the BLT reconstruction is ill-posed and underdetermined, with fewer measurements than unknowns. It is impossible to solve the equation by minimizing the discrepancy between the measurements ${\varphi _\mathrm{\lambda }}$ and the predicted values calculated from ${G_\mathrm{\lambda }}S$. Therefore, we apply the minimization equation M
In Eq. (4), ${\varphi _M}$ and ${\varphi _C}$ represent the optical intensity generated from the measurement and calculation, respectively. Here, ${\varphi _C}$ is determined by Eq. (3) and λ is the regularization parameter. In this research, we set λ is 100. L is the Laplacian structure, which is defined using US-derived priors and Δx is the difference of internal distribution of bioluminescence for each iteration.
In this study, to quantitatively assess the quality of the reconstruction results, we used the LE and the DICE [35,36]. The DICE indicates the morphological similarity between the reconstructed source region and the real tumor region in the US image. The region circled in the US image was used as the comparison reference. A high DICE indicates that the two regions have better similarities in both location and shape.
In Eq. (5), ${S_1}$ represents the 3D BLT reconstruction result and ${S_2}$ corresponds to the tumor region highlighted in the US image.
The LE is the distance between the barycenter of the reconstructed source and that of the true anomaly. A lower LE indicates a better reconstruction. The LE is measured using the following function:
In Eq. (6), $S{C_{re}}$ and $S{C_{ac}}$ represent the barycenter coordinates of the reconstructed source and the actual source, respectively. ||•||2 is the operator of Euclidean distance.
2.7 Focused single-element US transducer
In this study, we used custom-made focused single-element US transducers for US imaging. The custom-made focused single-element US transducer was suitable for small animal imaging and provided reliable anatomical prior information. The piezoelectric material of the US transducer was lead zirconate titanate (PZT). The details of the transducer are shown in Fig. 3. Figure 3(a) illustrates the schematic of our ultrasonic piezoelectric transducer. Figure 3(b) shows a photograph of the US transducer. Figure 3(c) indicates that the focal length of the transducer was 15 mm, and the frequency of the transducer was 7.8 MHz, which means the focal length could fully cover the mouse. Furthermore, in collaboration with the motor system, this transducer allows for imaging of mice of different sizes and scanning depths. The resonance (fr) and antiresonance (fa) frequencies were measured using an impedance analyzer (Agilent 4294A). Based on these measurements, the electromechanical coupling coefficient (kt) was calculated to be 0.60 using the relevant equation. We then conducted wide bandwidth measurements by analyzing ultrasonic reflection from a steel block submerged 15 mm deep in water. After performing a Fast Fourier Transform via MATLAB, the -6 dB bandwidth was determined to be 102%. Finally, we used our homemade transducer to scan a copper wire with a diameter of 0.1 mm and determined the axial and lateral resolutions to be 0.2 and 0.3 mm, respectively. Figure 3(d) displays the electrical impedance and phase versus frequency of the transducer. Figure 3(e) shows the hydrophone measurements of the acoustic pressure produced by our US transducer.
Fig. 3. Details of a custom-made focused single-element US transducer: (a) cross-sectional view schematic, (b) appearance, (c) measured pulse-echo response (red line) and its frequency spectrum (dashed line), (d) measured electrical impedance and phase versus frequency, (e) measured sound field by hydrophone.
3. Results
3.1 Ultrasound imaging
The mice were placed flat on a custom-made animal holder, and a thin layer of standard US gel was applied to ensure proper coupling of the US transducer to the tissue. US imaging was then performed through the abdomen of a lightly anesthetized mouse in the supine position, using a single-element 7.8 MHz transducer. The transducer, positioned downward and in contact with the US gel, was scanned to acquire 3D wide-field scans with a step size of 0.5 mm along the y direction. Figure 4(a) shows several two-dimensional (2D) B-mode US images that were acquired by an US transducer and a motor system (n = 3). These individual images were then assembled, or “y-stacked,” to create an integrated 3D image, as seen in Fig. 4(b). The acquisition process involved multiple sweeps over a predefined region of interest (ROI), enveloping the entire abdomen.
Fig. 4. (a) Transversal plane US image in the tumor region acquired by our custom-made single element US transducers, (b) mouse model with 3 organs, (c) the y-stacked US image sequences. (scale bar: 10 mm)
3.2 3D bioluminescence tomography reconstruction
We carried out 3D BLT reconstruction to verify the capability and performance of the US-assisted 3D BLT method for in vivo animal studies (n = 3). Figure 5 displays the 3D BLT reconstruction of a tumor-bearing mouse, captured 30 minutes after D-luciferin injection. Figure 5(a) presents a comprehensive 3D US mesh of the abdominal region and the tumor in mice was also created using US imaging. The tumor measured approximately 10 mm in length and 5 mm in width. Figure 5(b) displays the fusion images, which include the white light image and the bioluminescent image of the tumor-bearing mouse from different imaging angles (0 °, 45 °, and -45 °), which were used for the reconstruction. We used a single-view bioluminescent image (0 °) and multi-view bioluminescent images to reconstruct the 3D BLT. The 3D BLT reconstruction calculation was performed using the corresponding US data as the prior information in the NIRFAST toolbox. Figure 5(c) provides a 3D view of the BLT reconstruction, incorporating prior anatomical information from 3D US images and sectional images in sagittal, coronal, and transverse views. The contour of the tumor is outlined by black lines for visual comparison. Successful tumor localization was achieved through prior anatomical information in BLT reconstruction. Figure 5(d) shows the 3D view of BLT reconstruction without structural prior information. For quantitative analysis of the in vivo experiments, the DICE was calculated for the 3D BLT reconstruction results and US information of the corresponding model. Figure 5(e) presents the corresponding quantitative results with 3D prior information. To validate the accuracy of the 3D reconstruction, the size and location of the reconstructed bioluminescence source were represented by its Full Width at Half Maximum (FWHM) and compared to the US-measured tumor location. In the case of 3D BLT reconstruction with prior information, the overlap measured by the DICE was 88.5%, with a locational error of 0.4 mm. Conversely, in the 3D BLT reconstruction without prior information, the overlap measured by the DICE was 57.2%, with a LE of 2.1 mm. Figure 5(f) presents the corresponding quantitative results with a single-view bioluminescent image (0 °) or multi-view bioluminescent images. In the case of 3D BLT reconstruction using single-view data, the overlap measured by the DICE was 75.3%, with a LE of 1.1 mm. However, when the same 3D BLT reconstruction was performed using two-view data, the overlap measured by the DICE increased to 83.7%, with a reduced LE of 0.8 mm. Conversely, when the 3D BLT reconstruction was performed using multi-view data, the overlap measured by the DICE further increased to 88.5%, with a significantly reduced LE of 0.4 mm. This work highlights the exceptional ability of our imaging system to accurately reconstruct a 3D bioluminescence emission model of tumors.
Fig. 5. 3D BLT Reconstruction results of the in vivo study. (a) 3D mesh of the entire body and tumor in mice, created using US imaging, (b) fused ambient and bioluminescence surface mapping images in different viewing angles (0 °, 45 °, and −45 °), (c) the 3D view, along with the sagittal, coronal, and transverse section images of BLT reconstruction results and merged images of BLT and US data, where the contour of the US highlighted region is outlined in black (see Visualization 1), (d) the 3D view, along with the sagittal, coronal, and transverse section images of BLT reconstruction results without the aid of US information (see Visualization 2), (e) the quantitative results without 3D prior information and with 3D prior information, (f) the quantitative results with different viewing angles and prior information. (scale bar: 10 mm)
3.3 Longitudinal optical imaging of tumor metabolism
BLT reconstruction was performed in a 4T1 mouse breast cancer model to monitor the metabolism of D-Luciferin in mice (n = 3). Figure 6(a) displays the 3D BLT reconstruction results, along with the merged images of BLT and US data, presented as several transverse section images at various time intervals (every 10 to 60 minutes). The DICE was 82.7%. Figure 6(b) illustrates the peak accumulation of bioluminescence probes in the tumor, which occurred 30 minutes after injection. For this study, D-Luciferin was administered to the mice via tail intravenous injection, at a concentration of 15 mg/cc and a volume of 200 µL per mouse. The ex vivo testing was conducted 1-hour post-injection, and the tumor and other organs were imaged using an EMCCD with epi-illumination geometry. To validate the tumor-specific localization of the observed bioluminescent signals in vivo, tumor and other organs, including the stomach, liver, spleen, pancreas, kidney, and intestines, were collected at the end of the study and imaged ex vivo for bioluminescence. Figure 6(c) showcases an epi-bioluminescence image of the tumor and organs from a mouse. The bioluminescence images revealed bioluminescent signals emitted by the tumor area, while no strong bioluminescent signals were detected in any other tissues or organs. The bioluminescent signal was consistently present within the tumor area and not observed in other organs. Figure 6(d) demonstrates the bioluminescence intensity collected from different organs at 1-hour post-injection, indicating high bioluminescent signals within the tumor region. Generally, a robust bioluminescence intensity in the tumor was observed at the 30-minute mark post-injection.
Fig. 6. Summary of the cross-sectional view of 3D BLT reconstructed tumor images, (a) at 1 min, 10 min, 20 min, 30 min, 40 min, 50 min, and 60 min after intravenous injection (scale bar: 10 mm), (b) average bioluminescence intensity in the prior region, (c) epi-fluorescence image of organs at 1-hour post-injection (scale bar: 10 mm), (d) bioluminescent agent metabolism at 1-hour post-injection.
4. Discussion
This research study introduces the development of a novel 3D BLT and US-integrated dual-modality imaging system (Fig. 1). The novelty is that we developed a rotational stage and used an US-aided image reconstruction algorithm with the multi-view bioluminescence images to provide a 3D BLT reconstruction. This method can provide both volumetric and tomographic representations. The primary objective of the study was to investigate the system's capability in quantitatively mapping the bioluminescence dose in biological samples and its real-time tracking capability for tumor tissue. Unlike conventional methods such as CT or MRI, this system utilizes an anatomical US imaging technique to visualize tumors, effectively addressing the challenging inverse problem of 3D BLT reconstruction in small animals. By directly fitting the nonlinear mapping relationship between surface bioluminescence and internal sources, the US-aided method eliminates the need for iteratively solving the inverse problem. Experimental results from phantoms and in vivo mouse models demonstrated the system's ability to obtain anatomical information of the tissue through US imaging while providing details about bioluminescence dose deposition in local tissue. Integrating BLT reconstruction with 360-degree projections and heterogeneous structure information in the tumor region yielded more accurate results than a homogeneous model. The accuracy increased from 57.2% to 88.5% as measured by the DICE. This anatomical information is of immense value as it significantly enhances tumor targeting accuracy and minimizes collateral damage to surrounding healthy tissues, ultimately leading to improved experimental outcomes. Furthermore, the naturally registered bioluminescence and US images facilitate continuous monitoring of tumor tissue, enabling near real-time analysis.
To the best of our knowledge, we have developed the first integrated 3D BLT/US dual-modality imaging system in a cost-effective manner. This system provides several significant benefits. Firstly, the system incorporates a rotational stage, allowing for the simultaneous scanning of both imaging techniques. This streamlined approach simplifies the merging of the two modalities and significantly reduces the time required for image processing. Secondly, the US imaging technique employed in the system delivers sufficient resolution and volume information, enabling accurate 3D BLT reconstruction of tumor images in nude mice. Thirdly, compared to other anatomical imaging methods like CT or MRI, including the US imaging modality in the system provides a more cost-effective solution. Fourthly, although micro-CT is commonly used in multimodal BLT systems to provide anatomical structure, segmenting soft tissue is challenging due to its low contrast in CT images without contrast enhancement.
Structural prior information (3D prior) plays a crucial role in the reconstruction process of 3D BLT. It is worth noting that accurate prior information is crucial for achieving optimal imaging performance. Any inaccuracies in the group's prior information can lead to a decrease in spatial resolution [37,38]. Figure 4 showcases a series of high-resolution 2D B-mode US images captured including the tumor region. These US images serve as the structural information required for transforming a mesh into the 3D BLT reconstruction. In this paper, US images were employed to facilitate the segmentation of soft tissue organs in mice. The resulting US data was used to establish a mapping relationship between the 3D physical space of the tumor region and the 3D image space of BLT in the mice, thus enabling evaluation of the BLT reconstruction. By incorporating US imaging, effective segmentation of organs such as contour, spine, and tumor can be achieved. This anatomical information proves invaluable for BLT reconstruction and helps address the ill-posedness of the inverse solution. Additionally, preliminary research on phantom and ex vivo 2D FDOT simulation has been successfully conducted [31,39]. However, the alignment process between FDOT and US images proved to be complex, requiring manual adjustments of the imaging angle to reduce the time required for image processing. Therefore, our proposed automatic optics-US 360-degree scanning system offers tumor localization with a location error of 0.4 mm and a dice coefficient of 88%. Furthermore, registering optical and US data is enabled to be more convenient and accurate in our dual-modality imaging system.
Given the ill-posed nature of BLT reconstruction and the sparsity of the source distribution, researchers have proposed various reconstruction algorithms combined with different types of prior information [40]. In our approach, the assumption is made that the tumor region exhibits heterogeneity, as evidenced by the surface bioluminescence images. These images, captured from 360-degree projections, are then utilized in conjunction with the mesh surface derived from the US images for BLT reconstruction. It is important to note that outside of the tumor area, the assumption of homogeneity is made, which may impact the accuracy of the reconstruction results. Figure 5 showcases the results of the 3D BLT reconstruction. A comparative analysis was conducted between the reconstruction utilizing the 3D US structural prior and the reconstruction without any prior information. The anatomical maps were able to provide structural information for the reconstruction of the source. As depicted in Fig. 5(c), the BLT results exhibit excellent alignment with the US-highlighted regions. Figure 5(e) clearly demonstrates that incorporating 3D anatomical prior information in BLT significantly improves tumor resolution and positional accuracy. The addition of prior structural information resulted in an improvement in location accuracy, with a DICE of 88.5% and a LE of 0.4 mm. When the 3D BLT is without prior information, the corresponding DICE is 57.2% and a LE of 2.1 mm. Furthermore, Fig. 5(f) indicates a 13% improvement in the DICE and a reduction of approximately 0.7 mm in LE when transitioning from single view to multi-view reconstructions. Multi-view bioluminescence images provide more surface bioluminescence information in the tumor area than a single-view bioluminescence image, resulting in a more accurate approximation of surface bioluminescence distribution for 3D BLT reconstruction. Additionally, multi-view bioluminescence images give a more precise estimation of the optical diffusion from the bioluminescence source to a curved surface than a single-view bioluminescence image. This effect is particularly noticeable in a source that is blurred from the top view, such as a deeper source, as expected. This is because the multi-view data aims to enhance access to different parts of the tumor region. Therefore, 3D BLT is beneficial for monitoring tumor growth and guiding cancer treatment.
The proposed method for detecting tumors has been validated through in vivo experiments, demonstrating its feasibility and advantages. Evaluation of the accuracy of the BLT algorithm was conducted by analyzing the results of 3D BLT in an in vivo metabolism experiment, as well as the epi-bioluminescence of mouse tissues/organs (Fig. 6). The highest bioluminescence intensity at the tumor area occurred 30 minutes after injection, aligning with both the 3D BLT result and the mouse tissue epi-bioluminescence image. Notably, precise bioluminescence distribution was observed in Luc-4T1 tumor mice, with a DICE of 82.7%. Additionally, the surface bioluminescence image of the mouse tissue revealed a concentrated presence of D-Luciferin in the tumor area, suggesting its potential as an excellent bioluminescent probe for tumor detection. The use of the 3D BLT technique simplifies the identification of tumor areas and offers a superior option for real-time, longitudinal, non-invasive, monitoring of tumor biology within a living organism.
The proposed method has successfully achieved tumor localization and drug metabolism monitoring. Several modifications can be further implemented to improve results to enhance the outcomes of this study. Firstly, optimizing the imaging system's performance is critical. This can be achieved by utilizing BLT in a broader wavelength region, such as the Near-Infrared (NIR) region, which mitigates the effects of light scattering. This adjustment can enhance reconstruction resolution, DICE, and LE. Secondly, addressing the challenge of defining organ contours, such as the liver or kidney, in US images is crucial. This can be accomplished by implementing compounding and multi-focal imaging techniques to improve US image quality. Consequently, the improved organ segmentation might provide more information for the process of heterogeneous BLT reconstruction, thereby enhancing the accuracy of the reconstruction. The third issue is the interference caused by breathing movements during scanning. One possible solution is to utilize a homogenization algorithm for calibration purposes. The fourth issue is that manual segmentation is time-consuming and prone to errors. Therefore, we plan to develop an automatic segmentation method for circling segments in US images in our future work. As part of our future work, we will focus on resolving these challenges to further enhance the accuracy of 3D BLT reconstruction.
5. Conclusion
We developed an innovative low-cost 3D dual-modality imaging system with a homemade US system to achieve 3D BLT. The primary goal of this system is to use US-assisted BLT reconstruction techniques for registering small animal tumors in cancer research. Our dual-modality system accomplished integrated scanning using a co-registered scanning device. By leveraging single-element US imaging, we achieved cost-effective imaging results as the 3D prior information for achieving high accuracy of our 3D BLT reconstruction with a DICE of 88.5% and a localization error of 0.4 mm. In conclusion, our new system provides a reliable functional and anatomical imaging platform, enabling preclinical studies on BLT and further advancing the exploration of optical molecular tomography.
Funding
National Science and Technology Council (NSTC109-2221-E-010-001-MY3).
Acknowledgments
This work was supported by the National Science and Technology Council of Taiwan (Grant No. NSTC109-2221-E-010-001-MY3).
Disclosures
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. K. Butterworth, K. Prise, and F. Verhaegen, “Small animal image-guided radiotherapy: status, considerations and potential for translational impact,” Br. J. Radiol. 88(1045), 20140634 (2015). [CrossRef]
2. C. Darne, Y. Lu, and E. M. Sevick-Muraca, “Small animal fluorescence and bioluminescence tomography: a review of approaches, algorithms and technology update,” Phys. Med. Biol. 59(1), R1–R64 (2014). [CrossRef]
3. G. C. Kagadis, G. Loudos, K. Katsanos, et al., “In vivo small animal imaging: current status and future prospects,” Med Phys. 37(12), 6421–6442 (2010). [CrossRef]
4. K. O’Neill, S. K. Lyons, W. M. Gallagher, et al., “Bioluminescent imaging: a critical tool in pre-clinical oncology research,” J. Pathology 220(3), 317–327 (2010). [CrossRef]
5. J. Zhang, D. Chen, J. Liang, et al., “Incorporating MRI structural information into bioluminescence tomography: system, heterogeneous reconstruction and in vivo quantification,” Biomed. Opt. Express 5(6), 1861–1876 (2014). [CrossRef]
6. J. Shi, T. S. Udayakumar, Z. Wang, et al., “Optical molecular imaging-guided radiation therapy part 2: Integrated x-ray and fluorescence molecular tomography,” Med Phys. 44(9), 4795–4803 (2017). [CrossRef]
7. J. Shi, T. S. Udayakumar, Z. Wang, et al., “Optical molecular imaging-guided radiation therapy part 1: Integrated x-ray and bioluminescence tomography,” Med Phys. 44(9), 4786–4794 (2017). [CrossRef]
8. S. Yahyanejad, P. V. Granton, N. G. Lieuwes, et al., “Complementary use of bioluminescence imaging and contrast-enhanced micro—computed tomography in an orthotopic brain tumor model,” Mol. Imaging 13(4), 7290.2014.00038 (2014). [CrossRef]
9. J. A. Guggenheim, H. R. Basevi, I. B. Styles, et al., “Bioluminescence tomography improves quantitative accuracy for pre-clinical imaging,” in European Conference on Biomedical Optics (Optica Publishing Group, 2013), p. 87990 G.
10. G. Zambito, C. Chawda, and L. Mezzanotte, “Emerging tools for bioluminescence imaging,” Curr. Opin. Chem. Biol. 63, 86–94 (2021). [CrossRef]
11. S. K. Lyons, “Imaging mouse models of cancer,” J. Cancer 21(3), 152–164 (2015). [CrossRef]
12. X. Huang, J. Chisholm, J. Zhuang, et al., “Protein nanocages that penetrate airway mucus and tumor tissue,” Proc. Natl. Acad. Sci. 114(32), E6595–E6602 (2017). [CrossRef]
13. R. T. Sadikot and T. S. Blackwell, “Bioluminescence imaging,” Proc. Am. Thorac. Soc. 2(6), 537–540 (2005). [CrossRef]
14. B. Zhang, W. Yin, H. Liu, et al., “Bioluminescence tomography with structural information estimated via statistical mouse atlas registration,” Biomed. Opt. Express 9(8), 3544–3558 (2018). [CrossRef]
15. J. Liu, Y. Wang, X. Qu, et al., “In vivo quantitative bioluminescence tomography using heterogeneous and homogeneous mouse models,” Opt. Express 18(12), 13102–13113 (2010). [CrossRef]
16. G. Wang, W. Cong, K. Durairaj, et al., “In vivo mouse studies with bioluminescence tomography,” Opt. Express 14(17), 7801–7809 (2006). [CrossRef]
17. W. Cong, G. Wang, D. Kumar, et al., “Practical reconstruction method for bioluminescence tomography,” Opt. Express 13(18), 6756–6771 (2005). [CrossRef]
18. M. Chehade, A. K. Srivastava, and J. W. Bulte, “Co-registration of bioluminescence tomography, computed tomography, and magnetic resonance imaging for multimodal in vivo stem cell tracking,” Tomography 2(2), 158–165 (2016). [CrossRef]
19. P. Wu, K. Liu, Q. Zhang, et al., “Detection of mouse liver cancer via a parallel iterative shrinkage method in hybrid optical/microcomputed tomography imaging,” J. Biomed. Opt. 17(12), 126012 (2012). [CrossRef]
20. H. Yan, Y. Lin, W. C. Barber, et al., “A gantry-based tri-modality system for bioluminescence tomography,” Rev. Sci. Instrum. 83(4), 1 (2012). [CrossRef]
21. A. Ale, V. Ermolayev, E. Herzog, et al., “FMT-XCT: in vivo animal studies with hybrid fluorescence molecular tomography–X-ray computed tomography,” Nat. Methods 9(6), 615–620 (2012). [CrossRef]
22. B. Li, F. Maafi, R. Berti, et al., “Hybrid FMT-MRI applied to in vivo atherosclerosis imaging,” Biomed. Opt. Express 5(5), 1664–1676 (2014). [CrossRef]
23. T. J. Czernuszewicz, V. Papadopoulou, J. D. Rojas, et al., “A new preclinical ultrasound platform for widefield 3D imaging of rodents,” Rev. Sci. Instrum. 89(7), 1 (2018). [CrossRef]
24. J. D. Rojas, J. B. Joiner, B. Velasco, et al., “Validation of a combined ultrasound and bioluminescence imaging system with magnetic resonance imaging in orthotopic pancreatic murine tumors,” Sci. Rep. 12(1), 102 (2022). [CrossRef]
25. S. Choi, J. Y. Kim, H. G. Lim, et al., “Versatile single-element ultrasound imaging platform using a water-proofed MEMS scanner for animals and humans,” Sci. Rep. 10(1), 6544 (2020). [CrossRef]
26. K. J. Francis, R. Booijink, R. Bansal, et al., “Tomographic ultrasound and led-based photoacoustic system for preclinical imaging,” Sensors 20(10), 2793 (2020). [CrossRef]
27. C.-W. Wei, T.-M. Nguyen, J. Xia, et al., “Real-time integrated photoacoustic and ultrasound (PAUS) imaging system to guide interventional procedures: ex vivo study,” IEEE Trans. Ultrason., Ferroelect., Freq. Contr. 62(2), 319–328 (2015). [CrossRef]
28. G. Bal and J. C. Schotland, “Ultrasound-modulated bioluminescence tomography,” Phys. Rev. E 89(3), 031201 (2014). [CrossRef]
29. R. Liu, J. Tang, Y. Xu, et al., “Bioluminescence imaging of inflammation in vivo based on bioluminescence and fluorescence resonance energy transfer using nanobubble ultrasound contrast agent,” ACS Nano 13(5), 5124–5132 (2019). [CrossRef]
30. F. Chung, T. Yang, and Y. Yang, “Ultrasound modulated bioluminescence tomography with a single optical measurement,” Inverse Probl. 37(1), 015004 (2021). [CrossRef]
31. P.-A. Lo, S.-P. Su, and H. K. Chiang, “Small-animal 360-deg fluorescence diffuse optical tomography using structural prior information from ultrasound imaging,” J. Biomed. Opt. 25(03), 1 (2020). [CrossRef]
32. Y.-F. Chen, K.-W. Chang, I.-T. Yang, et al., “Establishment of syngeneic murine model for oral cancer therapy,” Oral Oncol. 95, 194–201 (2019). [CrossRef]
33. H. Dehghani, M. E. Eames, P. K. Yalavarthy, et al., “Near infrared optical tomography using NIRFAST: Algorithm for numerical model and image reconstruction,” Commun. Numer. Meth. Engng. 25(6), 711–732 (2009). [CrossRef]
34. B. Zhang, K. K.-H. Wang, J. Yu, et al., “Bioluminescence tomography–Guided radiation therapy for preclinical research,” Int. J. Radiat. Oncol. Biol. Phys. 94(5), 1144–1153 (2016). [CrossRef]
35. S. Jiang, J. Liu, G. Zhang, et al., “Reconstruction of fluorescence molecular tomography via a fused LASSO method based on group sparsity prior,” IEEE Trans. Biomed. Eng. 66(5), 1361–1371 (2019). [CrossRef]
36. J. Yu, C. Dai, X. He, et al., “Bioluminescence tomography based on one-dimensional convolutional neural networks,” Front. Oncol. 11, 760689 (2021). [CrossRef]
37. Y. Song, L. Shen, L. Xing, et al., “Solving inverse problems in medical imaging with score-based generative models,” arXiv, arXiv:2111.08005 (2021). [CrossRef]
38. B. Li, M. Abran, C. Matteau-Pelletier, et al., “Low-cost three-dimensional imaging system combining fluorescence and ultrasound,” J. Biomed. Opt. 16(12), 126010 (2011). [CrossRef]
39. P. A. Lo and H. K. Chiang, “Three-dimensional fluorescence diffuse optical tomography using the adaptive spatial prior approach,” J. Med. Biol. Eng. 39(6), 827–834 (2019). [CrossRef]
40. X. Chen, D. Yang, Q. Zhang, et al., “L1/2 regularization based numerical method for effective reconstruction of bioluminescence tomography,” J. Appl. Phys. 115(18), 1 (2014). [CrossRef]
