Sensitivity analysis aimed at blood vessels detection using interstitial optical tomography during brain needle biopsy procedures

1. Introduction

Brain tumors represent only ~1.5% of all new cancer cases but the incidence is increasing and the mortality is extremely high. The standard-of-care for most high-grade tumors is open-cranium surgery to achieve maximum resection, which markedly impacts survival, followed by radiation and chemotherapy [1–4]. The decision to resect is informed by the suspected tumor type and grade, which are roughly assessed by pre-operative magnetic resonance imaging (MRI) or computed tomography (CT), and by risk/benefit of the likelihood of permanent neurological deficit for lesions located in eloquent structures. When precise neuropathological diagnosis is required, a stereotactic brain needle biopsy (BNB) is performed to collect tissue prior to or in place of surgery, providing information (tumor type, grade, growth pattern) with greater sensitivity and specificity than imaging [1,5–8]. The main indications for BNB are deep-seated lesions, multiple lesions and lesions in surgical candidates with poor health. The objective is to collect at least one tissue sample from the area of highest World Health Organization (WHO) grade.

A disposable biopsy needle (Fig. 1) is used, comprising internal and external cannulas, a needle stop and an aspirator tube. The needle has an outer diameter (O.D.) usually ∼2 mm and is inserted through a burr hole very slowly to reduce damage to blood vessels. Both cannulas have an open rectangular window (the ‘cutting window’) into which tissue is drawn by suction. The internal cannula is then rotated to cut the specimen, which is withdrawn for histopathology analysis. There is considerable variability in the number and specific manner in which biopsy samples are collected during BNB procedures. Generally, 1 to 5 or more needle passes (serial biopsies) are required to try to sample the most critical tissue. This variability is influenced by human as well as tumor physiology factors. If the tumor is very heterogeneous more samples can be collected, both along and around the planned needle path.

 

Fig. 1 (left) A commercial biopsy needle is composed of 2 hollow tubes, both with a rectangular hole close to their distal end. The outer tube is a cannula that is inserted into the brain while the inner tube can be slid in and out of the cannula. When the needle (with cannula and inner tube windows not aligned) has been inserted into the brain and has reached a location where tissue needs to be collected, the inner tube is rotated so that the two windows become aligned, exposing brain tissue to the interior of the inner tube. (right) A fibre optics tomography system can be integrated onto a commercial needle to prevent blood vessels from being clipped off during a brain biopsy procedure.

Download Full Size | PPT Slide | PDF

Position accuracy is critical to achieve the correct neuropathological diagnosis on all types of lesions for which BNB is performed. However, the current standard-of-care can lead to misdiagnosis because of: (1) samples being collected in non-diagnostic areas along the needle pathway (e.g., missing the capsule of a high-grade glioma and sampling only necrotic areas); and (2) complete geographic miss of a lesion due to registration inaccuracies and tissue shifts. For example, up to 17.4% of procedures yield non-representative samples due to inappropriate target definition, resulting in incorrect tumor typing/grade [6,9–11]. Frozen sections are sometimes performed during the procedure to reduce misdiagnosis but this is time consuming, costly and significantly less definitive and accurate than histopathology on stained sections, which takes several days. In addition, intracranial hemorrhage, the rate of which varies widely between reports (0.7-30.7% of cases) [12–14], contributes substantially to morbidity (up to 16.1%) [9,15] and mortality (up to 3.9%) [9,16]. This is due to rupturing of a significant blood vessel being aspirated into the cannula. Since the surgeon does not have access to blood vessel location during BNB, this risk is always present.

There are several possible approaches to address the unmet clinical need of minimizing the risk of significant bleeding during BNB, In particular we considered the use of optical coherence tomography (OCT) as an established, high-resolution optical imaging technique that could be integrated into a biopsy needle. OCT relies on back-reflected ballistic and near-ballistic photons to provide depth scans in tissue and is widely used, for example in ophthalmology. The major limitation is that the maximum imaging depth (or radial distance in the case of interstitial imaging) in highly scattering tissue such as the brain is typically <~1 mm, which is not sufficient to meet the specific clinical need in BNB guidance. Hence, we have developed a novel interstitial (non-coherent) optical tomography (iOT) technique that can be integrated into the neurosurgical workflow to mitigate the hemorrhage risk during BNB procedures by locating blood vessels up to 2 mm distant from the biopsy needle, which is the typical size of the biopsy and so represents the tissue at risk. The method relies on the detection of high-absorbing brain areas associated with a large concentration of hemoglobin. We hypothesize that these signatures can be used as surrogates for clinically-significant blood vessels, since white and gray matter are relatively low light absorbers. The novel iOT technique shares similarities with diffuse optical tomography (DOT) but with two main differences: (i) the interrogated tissue is exterior to the illumination-detection geometry (because of the interstitial nature of the method), and (ii) light propagation modeling requires a treatment valid in the sub-diffusive transport regime (because some source-detector pairs are separated by sub-millimeter distances). A proof-of-concept performed on a realistic brain tissue phantom demonstrated that interstitial optical tomography (iOT) can detect several 1 mm-diameter high-contrast absorbing objects located ≤2 mm from the needle [17].

Here, a detailed sensitivity analysis is presented in order to quantify the intrinsic detection capabilities of iOT using numerical simulations and experimental tissue-phantom data sets obtained with the imaging prototype introduced in [17]. In Section 2 methods are presented including a description of the imaging system and the tissue-mimicking phantoms used to emulate blood vessels in brain tissue, both experimentally and using light transport simulation. A method is proposed for intraoperative iOT data visualization that is increasing the imaging specificity through spectral fitting using the signature of oxygenated and deoxygenated haemoglobin. Results of the sensitivity analysis are presented in Section 3 including a direct comparison between the numerical and experimental experiments. A discussion of the results is presented in Section 4.

2. Methods

2.1 Imaging system and tissue phantom experiments

The optical probe used to evaluate the sensitivity of the iOT system has been introduced in [17] and is schematically depicted in Fig. 2(B). It has an outer diameter of 1.7 mm and comprises 24 optical fibres (100 μm core diameter, 120 μm outer diameter, N.A. = 0.22) grouped in pairs. Individual micro-prisms are glued to the tip of each fibre pair for side-detection or side-illumination at 90° angle. Each fibre is connected to a cable ending with an SMA connector; in total a bundle of 24 connectors exits the optical needle as schematically depicted in Fig. 2(A). Tissue illumination is achieved with a broadband light source (HL-2000, Ocean Optics) connected to a 1x16 multiplexer (MPM-2000, Ocean Optics) sequentially delivering light to the 12 outermost fibres. Another 1x16 multiplexer transfers optical signals to a high sensitivity portable spectrometer (QE65pro, Ocean Optics) for sequential detection through the innermost fibres for each illumination point. Light signals are not acquired for the detection fibre sharing the same micro-prism as the illumination fibre. A full tomographic data set consists of 132 spectra (12 illumination points x 11 detection points) between 400 and 850 nm with ∼1 nm spectral resolution. All hardware components are controlled with a custom LabView (National Instruments) program.

 

Fig. 2 (A) Experimental setup with the optical probe immersed in a brain tissue-simulating phantom. (B) Optical probe with an enlarged view of the tip showing the positioning of the illumination and detection fibres. (C) Contrast ratios at 600nm for a tissue phantom and normal brain: (left) carbon rod used to emulate blood vessels absorption in a fat emulsion liquid background, (right) in vivo brain image during open cranium surgery illustrating the optical contrast between cortex (gray matter) and blood vessels.

Download Full Size | PPT Slide | PDF

The experimental setup is shown in Fig. 2(A). Tissue-simulating phantoms were made to mimic tissue optical properties: (i) a highly scattering, weakly absorbing liquid representing brain matter, and (ii) highly absorbing, weakly scattering tubular structures (carbon rods) simulating blood vessels. We are hypothesizing that carbon rods can be used as surrogates for blood vessels because white and gray matter (excluding the complex network of veins and arteries) are relatively low light absorbers when compared to haemoglobin. This is supported by Fig. 2(C) demonstrating that the apparent optical contrast at 600 nm of a 500 μm O.D. carbon rod is similar to that of blood vessels imaged during a human glioma resection procedure using a wide-field hyperspectral camera system (see [18] for imaging system details). The bulk medium was created with a fat emulsion (Intralipid 20% 1:5 V/V) and a blue coloring dye (Club House) diluted in water (1:1000 V/V) to obtain reduced scattering coefficient (μ_s’) and absorption (μ_a) values consistent with brain matter [19–21]. Figure 3 shows the absorption spectrum of the blue dye @ 1:1000 V/V where annotations are made next to the absorption coefficient values considered for the sensitivity analysis (see section 2.2.2), i.e., μ_a = 0.370 mm⁻¹, 0.309 mm⁻¹, 0.247 mm⁻¹, 0.186 mm⁻¹, 0.124 mm⁻¹, 0.063 mm⁻¹, 0.001 mm⁻¹ at respectively 631 nm, 640 nm, 645 nm, 650 nm, 655 nm, 663 nm and 700 nm: these coefficients are representative of the bulk absorption values expected in the brain (0.001 mm⁻¹ to 0.1 mm⁻¹) [21]. The absorption coefficient of the food coloring is measured using a custom spectrophotometer including a white LED source (MCWHL5, Thorlabs), a fiber optic coupled cuvette holder (CUV-UV, Ocean Optics) and a spectrometer (Maya 2000 Pro, Ocean Optics). From the measured intensity through 1 cm of water, the absorption coefficient is derived using Beer-Lambert’s law. The average reduced scattering coefficient of Intralipid 20% diluted to 1:5 V/V is μ_s’ = 6.77 mm⁻¹ and is constant over the spectral detection range of the iOT system with a standard deviation <2%.

 

Fig. 3 Absorption spectrum of the blue coloring dye (@ 1:1000 V/V) used to fabricate tissue-mimicking phantoms.

Download Full Size | PPT Slide | PDF

Heterogeneities were modeled with carbon rods of different O.D. placed at different positions relative to the probe. The distance between the rod and the probe are measured edge-to-edge, with a distance of 0 mm corresponding to the two objects in contact. As illustrated in Fig. 2(A), the probe is mounted on a high precision rotation mount (PRM05, Thorlabs) for precise rod alignment. The mount is affixed on a 3-axis translation stage (PT3, Thorlabs), which allows 25mm travel paths in all three directions for positioning the probe relative to the rod with <10 μm error. Each carbon rod is also fixed onto its own removable support. Overall, 51 hyperspectral iOT measurements were made with a single carbon rod immersed into the bulk diffusive medium. The rod O.D. were: 200 μm, 720 μm and 1350 μm, and the edge-to-edge distances were: 0 μm, 20 μm, 50 μm, 100 μm, 150 μm, 200 μm, 300 μm, 400 μm, 500 μm, 600 μm, 700 μm, 800 μm, 900 μm, 1000 μm, 1250 μm, 1500 μm, 2000 μm. For each configuration, an optical data set was also acquired with no blue dye present in the bulk diffusive medium. For all measurements, the detection time (for each source-detector pair) ranged from 18 ms to 5 s (with larger source-detector separations requiring more time) and was adjusted to consistently insure a noise level <2%. As a result, a full tomographic data set took approximately 6 minutes to acquire. However, it should be noted that several hardware modifications could be made to the system in order to take the acquisition time down to a fraction of a second. These could include (i) using a more powerful white-light, (ii) using a spectral camera instead of multiplexers in order to minimize the light losses and insure data from multiple detectors can be acquired simultaneously.

2.2 Brain tissue light transport numerical simulations

2.2.1 Monte Carlo simulation technique

Monte Carlo simulations (MCS) were performed to more fully evaluate the absorbers sensitivity of the system by including a broader range of parameters (e.g., inclusion size and bulk optical propoerties) that those considered in the experiments of Section 2.1. Full tomographic data sets were obtained by simulating the different laser injection and photon detection regions, each with pre-defined areas and numerical apertures. The optical needle geometry was modeled using a tetrahedral mesh (Fig. 4(A)) and simulations performed using Mesh-based Monte Carlo (MMC) [22,23]. The imaging geometry considered was that of the experimental system described Section 2.1 (12 sources, 11 detectors: 132 measurements in total; see Fig. 5(B)). A denser detection geometry with 36 optodes equally distributed on the periphery of the needle was also simulated (Fig. 5(C)). Simulations were conducted on an Intel(R) Core(TM) i7-4820K CPU @ 3.7 GHz desktop with 32 GB of RAM or the compute cluster Mammouth Parallel II, part of Calcul Canada network [24]. The mesh for the simulations was cylindrical (radius 40 mm, height 80 mm) and consisted of 196,729 nodes forming 1,212,849 tetrahedral elements. The extended light source (Fig. 4(B)) was modeled by 126 punctual sub-sources covering a circular area on the needle of ∼0.079 mm² consistent with the diameter of the optical fibres in the actual imaging system described in Section 2.1.

 

Fig. 4 (A) Mesh used for light transport simualtions in the medium surrounding the needle; detectors and sources are represented by spheres. (B) Modeling of the extended sources using a finite number of sub-sources.

Download Full Size | PPT Slide | PDF

 

Fig. 5 (A) Schematic depiction of the cylindrical absorber configurations simulated: 7 diameters were considered each associated with 10 edge-to-edge distances (70 configurations in total). (B) Spatial distribution of the optodes for the experimental iOT system, (C) Spatial distribution of the optodes for a different imaging geometry.

Download Full Size | PPT Slide | PDF

Each MCS (for each source-detector pair and the associated large number of injected photons) outputs the partial path lengths (PPL) of a detected photon packet. The PPL represents the distance travelled in each sub-region (each sub-region can have different optical properties) of the tetrahedral mesh for a packet, for which the attenuation can be computed with the Beer-Lambert law. The total measured intensity within the detection area is the sum of the energies associated with each photon packet:

Table 1. Optical properties considered for bulk tissue in numerical simulations.

View Table | View all tables in this article

These properties are representative of the different brain tissue optical properties found in the literature [20]. However, O.P. #4 (large absorption and large scattering) is likely unrealistic for brain tissue but was considered in order to test the sensitivity in a more extreme attenuating environment. In summary, O.P. #1, O.P. #2 and O.P. #3 likely capture the maximum attenuation values in brain when tissue excitation is in the near-infrared. In fact, intraoperative in vivo reflectance measurements during glioma surgery demonstrated that the reduced scattering coefficient of normal brain is approximately μ_s’ = 1.5 to 2.5 mm⁻¹ and the absorption coefficient lies between μ_a = 0.1 mm⁻¹ and μ_a = 0.001 mm⁻¹ above 600 nm [21].

2.2.2 Computation of sensitivity to absorbers metric

The sensitivity of the system to absorbers was evaluated based on the experimental measurements made on phantoms (Section 2.1) and Monte Carlo simulations. For the later, 7 cylindrical inclusion diameters were considered each associated with 10 different radial positions, as depicted in Fig. 5(A). The O.D. of the absorbers were 212 μm, 301 μm, 425 μm, 601 μm, 850 μm, 1.202 mm, 1.7 mm. The needle-absorber edge-to-edge distances were 0 mm, 0.23 mm, 0.5 mm, 0.82 mm, 1.22 mm, 1.71 mm, 2.3 mm, 3.03 mm, 3.92 mm, 5 mm.

Simulations were performed for the projections associated with 11 detectors as in Fig. 5(B). However, in order to evaluate the sensitivity to absorbers, a single laser source was considered. Namely, a source associated with the angular position that is facing the absorber directly. In other words, only measurements associated with the source closest to the inclusion were considered. In order to compute the sensitivity, the two detectors closest to the source were considered, namely the detectors that are at 30° and −30° from the source resulting in a source-detector distance of ∼445 μm. Simulations were performed for the four sets of bulk optical properties found in Table 1. The intrinsic reduced scattering coefficient of the cylindrical inclusion was set to μ_s’ = 0.05 mm⁻¹ under the hypothesis that most of the attenuation in vessels is associated with the absorption from haemoglobin molecules. The absorption coefficient of the inclusion was varied from the background value in Table 1 up to an unrealistically large value of μ_a = 5 mm⁻¹. A simulation was also performed associated with an homogenous background (no cylindrical inclusion) for all four sets of optical properties. A statistically relevant sensitivity study required large amount of photons: for example, each distance-diameter combination required 500 million photons for a simulation time of about 1 hour using a single node on the Mammouth Parallel II computer cluster (24 cores).

The metric that was used for computing the sensitivity of the system (in %, either experimentally or based on numerical simulations) is based on the formula

2.2.3 Two-dimensional representation of iOT data sets and spectral fitting

Angiography is an X-ray technique using a contrast agent and a fluoroscope to image blood flow in arteries and veins. Numerical Monte Carlo simulations were made on a mesh generated from the segmentation of the porcine brain angiogram shown in Fig. 6(A). The resulting mesh is shown in Fig. 6(B). It is composed of 128,147 nodes, 775,287 tetrahedral elements and has a volume of 50 x 40 x 70 mm³. As depicted in the figure, the biopsy needle is positioned in the center of the {x, y} plane and is moved along the z-axis for the purpose of generating iOT simulations at different depths. In total, 24 measurement slices (with 36 optical projections each as in Fig. 5(C)) were simulated from z = 0 to 30 mm with 1.25 mm steps. The simulations were performed for bulk optical properties O.P. #3 (Table 1). Blood vessels were separated into veins (blue) and arteries (red). The extinction coefficients of oxygenated haemoglobin (HbO) and deoxygenated haemoglobin (HbR) were used to assign optical properties for arteries and veins. Arteries were assigned an HbO concentration of 150 g/liter (with no HbR), and veins an HbR concentration of 150 g/liter (with no HbO). Both arteries and veins were assigned a reduced scattering value of μ_s’ = 0.5 mm⁻¹. The iOT numerical simulations were obtained for all wavelengths between 400 nm and 900 nm using the method described in Section 2.1.1. Each excitation point in tomographic data set took ~5 minutes (10 million photons) on a single node on the compute cluster: using 24 nodes in parallel (one per slice) kept the computational time to 3 hours when using 36 fibers.

 

Fig. 6 (A) CT angiogram of porcine brain showing the complex architecture of vessels and arteries. (B) 3-D images showing a segmentation of the angiogram with a threshold allowing to highlight the vessels and the generating of a mesh. The separation between arteries (in red) and veins (in blue) has been made for the simulations but does not reflect the actual physiology.

Download Full Size | PPT Slide | PDF

Optical sinograms at 600 nm were formed for each tomographic slice. The notion of optical sinograms and how they are computed were introduced in [17]. Briefly, they are a two-dimensional representation of an iOT data set where each projection (x-axis) is associated with the intensities of one light source and multiple detectors (y-axis). The sinogram is normalized to the maximum intensity value that is contains. Since no measurements are usually acquired for the detection fibre sharing the same position as the source fibre, the diagonal does not contain any data.

A linear least-squares regression is used to project each spectroscopic measurement of the 3D iOT data set onto the basis spectra for HbO and HbR [25]. A more complete treatment would require scattering to be considered as a source of optical contrast but for simplicity here the assumption was made that the majority of contrast was associated with absorption. Keeping only the sum of the measurements adjacent to each source position, a map was formed representing the relative contribution of HbO and HbR to the signal as a function of z and angular position around the biopsy needle in order to assess the possibility of using these simple maps as a mean to specifically detect blood vessels using more specific optical sinograms. For each source position around the needle, the variation in signal between homogeneous and heterogeneous was averaged for the nearest detectors. This is computed with the mathematical operation log₁₀ ratio $\left(-\log 10\frac{I_o}{I}\right)$. A linear fit of HbR/HbO spectra was performed to recover their relative concentration. Two color maps (cyan and magenta) were used to show the recovered HbR and HbO concentrations, respectively.

3. Results

Here results are presented of the analysis designed to evaluate the feasibility of using iOT for blood vessels detection during brain biopsy procedures. Specifically, the sensitivity S in Eq. (2) was computed using data sets generated from numerical simulations (Section 3.1) as well as obtained through tissue phantom experiments (Section 3.2). A preliminary analysis (based on numerical simulations only) is also presented in Section 3.3 relating to the multispectral functionality of the imaging system and its use to increase the specificity of the technique through spectral decomposition in terms of HbO and HbR. Visualization tools are also introduced in Section 3.3 whereby 2-D absorption or haemoglobin maps can be used as a surrogate for the presence of blood vessels during needle insertion into the brain.

3.1 Sensitivity analysis: numerical simulations

Figure 7 shows 12 sub-figures associated with simulations for 3 different cylindrical inclusion O.D. (212 μm, 601 μm, 1700 μm) and 4 different sets of bulk optical properties: O.P. #1, O.P. #2, O.P. #3, O.P. #4 from Table 1. Each graph represents the percentage of variation S computed from Eq. (1) as a function of the intrinsic absorption of the cylindrical inclusion, for 10 different edge-to-edge separations between the needle and the inclusion, ranging from 0 mm to 5 mm. On each sub-figure, the red line represents a realistic detection threshold that has been set to 2% based on the noise level associated with the experimental data acquired with the iOT system in Section 2.1.

 

Fig. 7 Detailed sensitivity plots based on the Monte Carlo simulations. Each sub-figure includes graphs (one for each needle-inclusion distance) of sensitivity, Eq. (2), as a function of intrinsic absorption of the cylindrical inclusion. Each column is associated with one set of bulk optical properties and each line is associated with a different diameter of the cylindrical inclusion.

Download Full Size | PPT Slide | PDF

In Fig. 7, whenever a curve is above the 2% noise level, it can be assumed that the inclusion can be detected. Inspection of the figure allows conclusions to be drawn relating to the intrinsic iOT detection limits in terms of inclusion size, optical contrast, distance-to-needle and bulk optical properties. In general, it is found that an inclusion with O.D. > 300 μm can be detected up to 2 mm from the needle for relatively low bulk absorption values (μ_a < 0.1 mm⁻¹), which is consistent with brain tissue above 660 nm [21]. As can be expected, larger inclusion O.D. and contrast values lead to larger values of S, and are thus more easily detected. Tables 2 and 3 show the sensitivity values for a subset of the data, for low- and high-absorbing bulk media respectively (O.P. #1 and O.P. #3), further supporting the predicted sensitivity limits of the iOT technique using numerical simulations.

Table 2. Sensitivity function S computed for different inclusion O.D. and edge-to-edge separations between the needle for a low-absorbing bulk medium. The absorption contrast between the inclusion and the bulk medium is 50000:1.

View Table | View all tables in this article

Table 3. Sensitivity function S computed for different inclusion O.D. and edge-to-edge separations between the needle for a high-absorbing bulk medium. The absorption contrast between the inclusion and the bulk medium is 50:1.

View Table | View all tables in this article

3.2 Sensitivity analysis: experimental evaluation

In order to reduce the amount of experimental measurements (compared to the exhaustive number of parameters considered for numerical simulations), the sensitivity analysis was reproduced experimentally using 3 carbon rods (O.D.: 310, 720 and 1350 μm) and blue dye was added in Intralipid to change the absorption over the spectroscopic detection range of the iOT system. Figure 8 shows acquired spectra without and with blue dye (left and right, respectively). The bottom graphs (Figs. 8(C) and 8(D)) show the difference in percent between homogeneous and heterogeneous media for different positions of the carbon rod. Figure 8(D) shows that this difference varies spectrally when blue dye is added because of the shape of its absorption spectrum (Fig. 3). Without the blue dye the difference is almost constant between 620 and 720 nm (Fig. 8(C)) because the Intralipid optical properties (scattering and absorption) are nearly unvarying over that range.

 

Fig. 8 (A) Spectra acquired for a single O.D. = 720 μm carbon rod immersed in a non-absorbing diffusive medium (no blue dye) for distances from the needle (edge-to-edge) ranging from 0 to 2 mm. (B) Spectra acquired for the same rod and edge-to-edge distance this time immersed in an absorbing diffusive medium (with blue dye). (C, D) Sensitivity metric S computed for each wavelength in the case of non-absorbing (in C) and absorbing bulk media (in D) with absorption values labeled for selected wavelengths highlighted with vertical lines (see Fig. 3).

Download Full Size | PPT Slide | PDF

Because of the non-trivial features of the blue dye absorption spectrum (Fig. 3), the sensitivity could be tested for various background absorption levels using a single spectroscopic measurement for each source-detector pair.. However, the absorption values considered in Fig. 8 are significantly larger than those considered in the numerical simulations, allowing the detection sensitivity limits of the system to be tested beyond the absorption range deemed relevant for brain tissue [21]. Inspection of Fig. 8 shows that detection of an O.D. = 720 μm rod can be achieved for all edge-to-edge separations (up to 2 mm) for media with absorption coefficients satisfying the condition μ_a < 0.15 mm⁻¹, which is consistent with assessment made based on the light transport simulations in Section 3.1.

In accordance with the numerical simulations, the impact on the spectroscopic measurements increases with diameter, reduces with distance and absorption, and high optical density (μ_a > 0.15 mm⁻¹) hinders the detectability of the carbon rods. In order to confirm consistency between experimental data and light transport simulations, MCS were performed for the exact same parameters as in the tissue phantom experiments for a O.D. = 720 μm rod. Figure 9 shows the resulting sensitivity function S for 4 different bulk absorption values ranging from μ_a = 0.001 mm⁻¹ to 0.370 mm⁻¹. We find that the simulation results are consistent with the experimental results, thus validating our MCS methodology and the conclusions reached based on their analysis.

 

Fig. 9 Comparing experimental and simulation results: inclusion O.D. = 720 μm, 4 different bulk absorptions (μ_a ranging from 0.001 to 0.370 mm⁻¹ as shown in the legend) and μ_s’ = 6.77 mm⁻¹. Experimental result are represented by full lines while MCS results are represented by single points (circles).

Download Full Size | PPT Slide | PDF

Figure 10 shows the experimental results for all three O.D. values, and for a large range of bulk absorption properties. An important conclusion to be drawn is that, in the case of low absorption (μ_a = 0.001 mm⁻¹), the carbon rods are detectable up to 2 mm from the optical probe for O.D. values as low as 310 μm when the threshold of detection is set at 2%. However, increasing bulk absorption can hinder detectability significantly: for example, the O.D. = 310 μm carbon rod is detectable for up to 2 mm only for μ_a < 0.001 mm⁻¹. If a detection threshold of 1% is considered instead, detection of that O.D. value up to 1.5 mm distance is possible for μ_a < 0.063 mm⁻¹, clearly demonstrating the possibility to enhance detection by diminishing the data noise level and optimizing data acquisition wavelengths to insure tissue is probed for wavelengths where brain optical properties are as small as possible.

 

Fig. 10 Left: Sensitivity metric S as a function of excitation wavelength for bulk media with different absorption coefficients. Right: Sensitivity for each bulk absorption values as a function of distance between the carbon rod and the optical probe: O.D.: 310 μm (A), 720 μm (B) and 1350 μm (C).

Download Full Size | PPT Slide | PDF

3.3 Two-dimensional representation of iOT data sets

Full tomographic data sets were computed using MCS to develop visualisation tools adapted for rapid blood vessels detection during surgery. One option consists in visualizing the data as 2D optical sinograms as in Fig. 11. These sinograms, shown at a specific wavelength in the figure, offer insight on the localization of absorbers around the biopsy needle although they do not provide depth-resolved information since no reconstruction algorthim was used. Figure 12 further demonstrates the impact of increasing the number of optical fibres when creating sinograms using MCS: the 36 fibres geometry (Fig. 5(C)) shows a significant improvement in sinogram quality compared to the 12 fibres model (Fig. 5(B)). As a result, it is expected that denser optode geometries will eventually result in improved reconstructed iOT images and will allow the detection of smaller absorbers.

 

Fig. 11 Four tomographic slices (top, arteries shown in red and veins in blue) and their optical sinograms @ 600 nm (bottom). Blood vessels close to the needle have a strong impact on the measurements and this can be observed on the optical sinograms: green and magenta circles show the blood vessels and their corresponding impacts on the sinogram.

Download Full Size | PPT Slide | PDF

 

Fig. 12 Blood vessels near the needle for slice #17 (left). Corresponding optical sinograms for 12 fibres (center) and 36 (right). Being in the sub-diffuse regime, an increase in quality is observable.

Download Full Size | PPT Slide | PDF

Data can also be visualized as a 2D map of HbR/HbO around the needle, as in Fig. 13. This allows specific blood vessels detection based on intrisinc optical contrast but can also be used for blood vessels detection using, e.g., exogenous dyes such as ICG.

 

Fig. 13 Using the HbR and HbO spectra to fit the simulated spectroscopic data can lead to specific blood vessels detection and separation from other potential sources of absorption. (A) Reconstructed map of HbR (cyan) and HbO (magenta). (B) HbR and HbO maps overlaid with actual blood vessels locations segmented from CT angiograms (arteries in red and veins in blue). (C) 3D representation of the data shown in (B) with 2D maps wrapped around the needle, segmented blood vessels from the CT angiogram are shown in red (arteries) and blue (veins).

Download Full Size | PPT Slide | PDF

4. Discussion

The hypothesis behind the work presented here is that the optical contrast (spatial and spectral) due to blood absorption can be used to detect significant blood vessels in the vicinity of a brain biopsy needle in order to avoid rupturing them during tissue extraction. Several techniques could potentially be used in order to exploit the optical contrast of vessels in order to increase the safety of these procedures. In optical microscopy (e.g., optical coherence tomography), the detected light is minimally scattered, giving very high spatial resolution (on the micrometer scale) but limited penetration depths of typically <1 mm. Conversely, diffuse optical tomography (DOT) techniques for large-volume imaging exploits diffused light, resulting in relatively poor resolution (~2-5 mm) but providing larger penetration depths (up to several centimeters) and the potential for high-specificity to haemoglobin detection when spectroscopic data is acquired.

In this work, we provided a detailed sensitivity analysis relating to the use of a sub-diffuse version of DOT for interstitial blood vessels detection. The short source-detector distances imposed by the needle geometry and the interstitial measurements lead to the development of short-range reconstruction algorithms, in which the sampling volume and resolution lie between those of microscopy and DOT [17]. Unlike conventional DOT, the imaging volume for interstitial optical tomography (iOT) is exterior to the sources and detectors whose separations are such that the diffusion approximation is not necessarily accurate, therefore requiring the use of Monte Carlo simulations for light transport modeling.

A probe is currently being developed by our group where the iOT system is integrated directly onto commercial biopsy needles in order to miminize the disruption of the surgical workflow by insuring that the optical measurements can be made during needle insertion prior to tissue cutting an extraction. In this work, however, a detailed sensitivity analysis was presented using a stand-alone iOT probe that was developed for proof-of-principle studies that the proposed tomographic geometry can be used for blood vessels detection [17]. In simulations, detectability is proven for absorber diameters as small as 212 μm located up to ∼2 mm from the biopsy needle core using a realistic instrumental noise level of 2% that is well in the range of usability for our application. Experimentally, it is demonstrated that absorbers (as small as 310 μm in diameters) can be detected up to 2 mm in the case of absorption values consistent with brain matter excluding tissue vasculature. For cases where bulk absorption is significantly higher, it is found that vessels could be difficult to detect for distances >1 mm, although the absorption values considered may have been too large to accurately model brain tissue excluding haemoglobin.

The ultimate goal of the proposed approach is to provide detection capabilities of significant blood vessels up to ∼2 mm from the needle since this corresponds to the maximum tissue sample size that can be aspirated into the needle prior to cutting. As a result, detection of those vessels can decrease the risk of hemorrhaging during the procedure. An important conclusion, based on both the numerical simulations and the tissue phantoms experiments, is that this can be achieved as long as the tissue optical properties (outside of the vessels that need to be detected) are small enough. However, a limitation of our study is that the spectral detection capabilities of the system were only tested numerically. Namely, the capabilities of the system to detect hemoglobin inclusions, and to further distinguigh beteween vessels containing oxygentated and deoxygenated vessels, was only assessed based on numerical simulations. In terms of the iOT technique itself, its capabilities are limited by the fact it likely will not be able to distinguish between a single blood vessel and a dense cluster of small capillaries. As a result, when used clinically this approach would guide the surgeon away from any region with high absorbance signal, whether it represents either of the cases. However, it is understood that the consequences associated with clipping a large blood vessel are worse than clipping a cluster of smaller blood vessels due to a decrease in both blood pressure and blood flow in the capillaries. This situation is an example of where it would possibly be preferable to use OCT since smaller vessels could be resolved. There is a clear trade-off between the two technologies. From a safety perspective, we believe that it would still be a greater risk to use OCT, as it could miss greater blood vessels located outside of its depth range. Further clarifying the usefulness of our technique for vessels detection and hemorrhaging prevention will be achieved by our group by testing and validating the approach in vivo using porcine brains.

It is also demonstrated (Fig. 11) that contrast in absorption can be observed directly on optical sinograms without the requirement for producing depth-resolved maps using a tomography algorithm, thereby rapidly providing surgeons with an estimation of where blood vessels are located. As demonstrated in Fig. 13, the information content of optical sinograms is even more substantial when multispectral data are used to exploit the signature of haemoglobin leading to semi-quantitative maps of HbR and HbO. However, it is expected that more accurate and robust depth localization will be achieved in vivo using iOT reconstruction algorithms currently being developed by our group.

Utlimately, the proposed technology will provide surgeons with new information to mitigate the risk of hemorrhage during BNB procedures. These improvements over current practice represent a significant potential clinical impact. In addition, the technology should contribute to improving health care cost-effectiveness by reducing inefficiencies and downstream costs of the standard BNB procedure in that it will decrease the number of repeat biopsy procedures due to misdiagnosis and reduce the number of frozen sections, which are expensive and time-consuming.