2.1.

Imaging System

2.1.1.

Optical brain biopsy needle

The diffuse reflectance fiber optics system is integrated directly onto the outer cannula of a commercial brain biopsy needle (model 9733068, Medtronic, Colorado), as shown in Fig. 1. The commercial brain biopsy needle is composed of two hollow tubes herein labeled as the inner and the outer cannulas (Fig. 1). The outer cannula is 20 cm in length with an outer diameter of 2 mm. At the distal end of each cannula, there is a small rectangular opening (henceforth referred to as the biopsy window) of dimensions $1.5\times 8\mathrm{mm}$ through which a sample can be aspirated, cut (through rapid rotation of the inner cannula relative to the outer cannula), and then harvested for pathology analyses. A requirement for the design of the optical biopsy needle is to ensure that the imaging system could be integrated directly on a commercial biopsy needle for minimal disruption of the surgical workflow. The integrated probe consists of 18 optical fibers (NA 0.22, Low-OH, $105\mu \mathrm{m}$ core diameter, FG105LCA, Thorlabs) disposed circumferentially around a 200-deg portion of the outer cannula of the commercial brain biopsy needle, leaving a blind spot (Fig. 1). The fibers are disposed opposite the biopsy window, and the angular range they cover is limited in order to leave the later unobstructed. As a result, imaging could be achieved on the opposite side of the window, and hence the need to rotate the probe by 180 deg between optical measurement and tissue collection. The probe design is compatible with the use of a standard stereotactic frame that is used for needle guidance, which allows to precisely control its rotation angle to ensure spatial registration between imaging and tissue-harvesting locations. A custom metallic angled aluminum mirror ($\sim 45\text{-}\mathrm{deg}$ inclination) is fixed at the tip of the probe to allow tissue illumination and light detection perpendicular to the needle. The optical fibers are glued with fast curing optical adhesive (NOA81, Thorlabs) on a thin open-ended hollow metallic tube (2.11-mm internal diameter and 2.24-mm external diameter) that could fit onto the external cannula of the commercial brain biopsy needle (2.11-mm external diameter). The diameter of the resulting assembly is 2.49 mm, including the external cannula, the hollow tube, and the fibers covered by a layer of optical glue for protection. The center-to-center distance between adjacent optical fibers is $\sim 100\mu \mathrm{m}$.

Fig. 1

Optical probe design showing the distribution of the optical fibers around the needle as well as the position of the mirror. Only the external cannula of the commercial needle is shown, with the optical fibers (in blue) disposed on its exterior surface: (a) transverse view and (b) longitudinal view of the optical needle shown in (c).

2.1.2.

Illumination/detection system

A fibered illumination/detection system was connected to the 18 optical fibers of the needle through SMA connectors (Fig. 2). A white light source (HL-2000 HP232, 450-1100 nm, Ocean Optics) was coupled to nine of the biopsy needle fibers using a mechanical optical switch (16-to-1 switch, MPM-2000, Ocean Optics). The illumination fibers (labeled with even numbers) alternated with the nine detection fibers on the biopsy needle (labeled with odd numbers). The latter were linked to another mechanical optical switch conducting the incoming light to a high-sensitivity spectrometer (QE65 Pro, Ocean Optics). The system was controlled using a custom Labview (National Instruments) program, automatically acquiring and storing reflectance spectra for all combinations of the illumination and detection fibers. As there were nine detection fibers and nine illumination fibers, 81 spectra in total could be acquired with the system. The acquisition time was the sum of the integration times for each spectrum plus an overall 500-ms delay associated with the motion of the mechanical components in the optical switches. In total, 13 mW was delivered to the distal end of the fibers for each illumination point.

Fig. 2

(a) Schematic representations of the optical system showing the white-light source and the spectrometer connected to mechanical fiber switches allowing a wide-angle interrogation when the needle is immersed in a tissue phantom composed of multiple inclusions (black carbon rods) embedded in a scattering medium (Intralipid™). (b) Geometrical test configuration around the optical biopsy needle where each of the nine potentially active fiber sources are identified with an even number, whereas the nine detection sources are identified with an odd number. Those numbers are associated with the numbering scheme used for identifying the optical switch positions in (a).

2.2.

Tissue Phantom Experiments

An experimental protocol was developed to test whether or not the new system was able to perform at the same level as the previous interstitial probe based on a stand-alone design that was not integrated onto a commercial biopsy needle.³⁶ To this end, optical phantoms were fabricated to reproduce the geometrical and optical properties of human brain under the assumption that absorption was mostly due to Hb, and that scattering was approximately two orders of magnitude larger than absorption.^38,39 The bulk medium associated with the diffusive phantoms was composed of a solution of Intralipid™ to emulate scattering (20%, 1:4 V/V) and blue coloring dye diluted in water for background absorption (1:1000 V/V). Two high-absorption carbon rods (outer diameters: $720\mu \mathrm{m}$ for inclusion 1, $310\mu \mathrm{m}$ for inclusion 2; see Fig. 2) were used to provide optical contrast to model blood vessels. The biopsy needle was affixed to a support, allowing its position in the bulk medium to be controlled using a three-axis translation stage (PT3, Thorlabs). Several parameters were allowed to vary in the experiments: (1) images were acquired with a varying distance between inclusion 1 and the needle, with a fixed distance between inclusion 2 and the probe, and (2) an image was acquired with the bulk medium composed only of Intralipid™ (no blue dye) as a reference for data normalization. For each dataset, the edge-to-edge distances between inclusion 2 and the optical biopsy needle were 0, 50, 100, 200, 500, 1000, and $2000\mu \mathrm{m}$. This resulted in the reconstruction of seven images using 81 reflectance spectra (i.e., all possible combinations of illumination/detection fibers) with an integration time for the reflectance spectra increasing from 1 to 5 s, with larger imaging times used for larger source-detector distances. The optimal integration time was estimated for each pair of fibers to maximize usage of the dynamical range of the sensor.

2.3.

Optical Reconstruction Algorithm

The reconstruction algorithm developed in Ref. 37 was used to image variations in optical absorption from a heterogeneous medium based on differences between data acquired in the heterogeneous medium (targeted for reconstruction) and calibration data acquired in a homogeneous medium. The technique was inspired from modeling approaches developed in diffuse optical tomography when solving the diffusion equation in the Rytov or Born approximation under the assumption that local variations in optical properties $({\mu }_a,{\mu }_s^{\prime })$ can be treated as perturbations.^40,41 Briefly, input data for each source/detector (S/D) pair associated with a spectral band of width $\mathrm{\Delta }\lambda$ (between wavelengths ${\lambda }_i$ and ${\lambda }_f$) was modeled using the following equation, where the approximation was made that the reduced scattering coefficient was constant across the imaging domain:

Physically, this equation was used to assign to each $S/D$ measurement a weight quantifying its relative contribution with respect to other measurements when contributing to variations in optical properties. Finally, a reconstructed image was computed by superposing the sensitivity function of all $S/D$ pairs, each weighted by the estimate of $\delta {\mu }_a^{S,D}$.

Reconstruction process can be achieved for any spectral band by using intensities in Eqs. (1) and (2) associated with the integral of the diffuse reflectance spectra over the desired spectral domain between ${\lambda }_i$ and ${\lambda }_f$. Integrals of the whole diffusion spectra were used for the tissue phantom experiments (Sec. 2.2) but a spectral band around 575 nm ($\mathrm{\Delta }\lambda =10\mathrm{nm}$) Hb peak was use for the in vivo experiments (see Sec. 2.4 below) to ensure that the images reconstructed are as specific as possible to blood vessels.^44,45 The calibration data (homogenous medium) used for the swine brain experiments consisted in reflectance spectra from an Intralipid™ and blue dye phantom designed to match as closely as possible the literature values for brain tissue optical properties (${\mu }_a=0.1{\mathrm{mm}}^{-1}$,^45,46 and ${\mu }_S^{\prime }\sim 8{\mathrm{mm}}^{-1}$).⁴⁶ To discriminate blood vessels from nonvascular tissue, we used Eq. (2), with $I_{S/D}^{\text{homogeneous}}$ as the intensity collected for the calibration phantom and $I_{S/D}^{\text{heterogeneous}}$ as the intensity from the interrogated medium. The weighted differences $\delta {\mu }_a^{S,D}$ were then tresholded, keeping only values at 50%, or higher, of the maximum value for the whole dataset. This threshold was a free parameter of the processing that we used to maximize the correlation between the reconstructed images and the x-ray images used for in vivo validation of the technique. A more sophisticated model, based on Monte Carlo simulations, was also developed to retrieve the optical absorption coefficient ${\mu }_a$ of the medium under the assumption of constant reduced scattering coefficient (${\mu }_s^{\prime }=8{\mathrm{mm}}^{-1}$ for brain tissue^36,38,39) Briefly, diffuse reflectance spectra were generated based on Monte Carlo simulations for a range of optical properties consistent with brain tissue. Then, using a least-squares algorithm the resulting spectral look-up table was used to associate absorption coefficient values for each wavelength of all experimental reflectance spectra. Using a curve fitting with a linear regression, each resulting ${\mu }_a$ spectrum was projected onto a basis composed of the extinction coefficients of oxygenated and deoxygenated Hb to retrieve their relative contributions. The resulting fitted Hb absorption spectra were used for image reconstruction only when ${\mu }_a>0.4{\mathrm{mm}}^{-1}$ in the 575-nm band because this threshold was found to most consistently reconstruct vessels detected with the gold standard of x-ray imaging. However the images reconstructed with this spectral method led to the same blood vessel detection assessment when compared to Fig. 6, and as a result they were not shown for conciseness.

2.4.

In Vivo Data Acquisition Protocol

To demonstrate the technical feasibility of using the optical brain biopsy system in an operating room, as well as its capacity for intraoperative detection of blood vessels, an in vivo experiment was designed and conducted during which it was used in a living adult swine (male, $\sim 30\mathrm{kg}$). In this experiment, the gold standard for blood vessel detection against which to evaluate the new technique was x-ray imaging. The animal ethics protocol was approved by the Comité institutionnel de protection des animaux at the Centre de recherche du Centre hospitalier de l’Université de Montréal (CRCHUM) in accordance with the Canadian Council on Animal Care guidelines. The animal was kept under general anesthesia during the entire experiment. The anesthesia was terminal.

Prior to the surgical procedure, the surgeon used two three-dimensional (3-D) imaging operating modes of the x-ray system (Artis Q, Siemens, Germany). The first was computed tomography (CT, 20 s acquisition per scan, 1 frame every 0.5 deg, 217 deg in total, $500\text{-}\mu \mathrm{m}$ resolution). The second imaging mode was angiography ($250\text{-}\mu \mathrm{m}$ resolution), obtained by subtracting two CT scans. For angiography, the second scan differed from the first because the contrast agent Iodixanol™ was injected ($270\mathrm{mg}/\mathrm{mL}$, $2\mathrm{mL}/\mathrm{s}$, and 50 mL) to highlight the blood vessels. The contrast agent was injected through a catheter into the femoral artery 5 s before image acquisition to ensure that it had spread into the arteries. Based on the imaging results (3-D image of the arteries), the surgeon planned a biopsy needle insertion trajectory to minimize the risks of damaging critical vessels. Although angiography had superior spatial resolution, CT scan images were shown to be more sensitive to vessels and thus were used to provide a postoperative reference in order to preliminarily assess the ability of the optical biopsy system to detect the presence of blood vessels.

During insertion of the biopsy needle, the fluoroscopy modality of the x-ray imaging system was used as a guide. Two two-dimensional (2-D) fluoroshots were saved at several points along the trajectory and combined to retrieve the 3-D location of the biopsy tool. At each imaging position, 30 reflectance spectra were acquired to reconstruct an image instead of the 81 used previously in the in vitro experiments. This constraint was imposed by the need to ensure that the full procedure was kept under 60 min as stipulated in the animal protocol. This resulted in an integration time of 55 s per image for each of the nine different depth locations along the trajectory of the optical biopsy needle. The spatial insertion step between two locations along the trajectory of the needle was $\sim 3\mathrm{mm}$. After the ninth acquisition, the target was reached covering a total insertion length of 23 mm from the initial depth. The optical biopsy needle was then removed. Each spatial position where optical data were acquired was then retrieved from the 2-D fluoroshots and merged with a CT scan. This resulted in a 3-D image map of blood vessels in the brain over which the trajectory of the needle was superposed. Image merging was done using the Medical Imaging Interaction Toolkit (MITK)⁴⁷ to provide reference data for blood vessel localization. Quantitative estimation of the registration error associated with disruption of the vessels is left for future work and remains here evaluated to $\sim 1$ to 2 mm.

3.1.

Tissue Phantom Experiment

The aim of the phantom experiments is to reproduce a subset of results obtained in Refs. 36 and 37 with a different probe design in order to demonstrate that the current system can also localize highly absorbing inclusions. For succinctness, images are presented only for a subset of the phantoms that are studied. Figure 3 shows reconstruction results of 4 out of the 28 phantoms that are imaged in this work. The figure shows images of carbon rods immersed in a diffusive medium with optical properties ${\mu }_a=0.01{\mathrm{mm}}^{-1}$ and ${\mu }_s^{\prime }=8{\mathrm{mm}}^{-1}$, with edge-to-edge distances between the optical biopsy needle and inclusion 1: (a) 0 mm, (b) 0.5 mm, (c) 1 mm, and (d) 2 mm. The other carbon rod (inclusion 2) remains at the same position in all four phantoms. For each image, the location of the carbon rods is indicated with black circles, whereas the color scale values (contrast) can be interpreted as being proportional to the probability that an absorber is present at the location of a given pixel. The cross section of the biopsy needle is represented with the white circle in the center of each image. The results in Fig. 3 demonstrate a match between the estimated angular location of the inclusion around the probe and its actual position for each of the four datasets. Furthermore, the decrease of the optical contrast when inclusion 1 is moved away from the edge of the probe shows the radial sensitivity of the system to be on the order of 2 mm from the edge of the probe. This is consistent with the radial sensitivity found in Refs. 36 and 37, and this finding is also supported by the reconstructed images (not shown for conciseness) of the other 24 phantoms with different background optical properties. Here depth sensitivity to inclusion 1 has been assessed based on whether any contrast could be observed in the radial direction from the needle to the inclusion. For example, although small for larger distances between inclusion 1 and the needle, contrast could still be observed up to an edge-to-edge separation of 2 mm despite the presence of a large contrast absorber next to the needle (inclusion 2). Each measured spectrum satisfies the requirement that its signal-to-noise ratio (SNR) is superior to 5, using the formula $\mathrm{SNR}=\sqrt{I}$, where $I$ is the number of photon counts in the 450- to 950-nm spectral band. For this experiment the SNR value estimated for each reflectance spectra is between 7 and 83, with an average value of 22 and a standard deviation of 18.

Fig. 3

Reconstructed diffuse optical images of the carbon rods immersed in a diffusive medium. The distances (edge to edge) between the optical biopsy needle and inclusion 1 are (a) 0 mm, (b) 0.5 mm, (c) 1 mm, and (d) 2 mm. The other carbon rod (inclusion 2) remains at the same position in all experiments.

3.2.

In Vivo Experiment

To evaluate the potential of the system to detect blood vessels in vivo, the 3-D CT image map is analyzed for use as the gold standard for vessels’ detection. Figure 4(a) shows a large field-of-view image of the CT volume, whereas Fig. 4(b) shows an enlarged view around the vicinity of the probe. In that enlarged view, the needle is shown in white, whereas the system holding it (made of a titanium tube and three screws) and the detected vessels are all shown in red. The large field-of-view image in the figure shows the swine skull along with the holding system and the needle. The reconstruction of the blood vessel network inside the brain provides the information necessary to estimate the location and the size of the vessels in the close vicinity of the needle. The precision for the vessels’ localization is limited by the error registration of the probe and the 3-D CT image map roughly estimated to 1 mm.

Fig. 4

CT scan used for locating blood vessels in the vicinity of the biopsy needle: (a) representation of blood vessels (red) surrounding the optical biopsy needle (gray tubular object) and (b) 3-D reconstruction of the surgical field in semitransparency showing the needle (in gray), the swine skull (in brown), and the blood vessels (in white).

Figure 5 shows two representative diffuse reflection spectra out of the 30 that are acquired per biopsy needle position, along with the calibration spectrum associated with a homogenous medium. Each spectrum in the figure shows characteristic features of the Hb absorption signature, including an increased absorption around 575 nm. The figure also shows the optical sinogram for one needle position, which graphically represents intensities associated with all 30 reflectance spectra with the detector number on the vertical axis and the source number on the horizontal axis. The intensity values in the sinogram are associated with the difference between the in vivo data and the calibration data (homogenous medium) at 575 nm. The sinogram data are used directly as input in Eq. (2) for image reconstruction. For this experiment, each measured spectrum used for image reconstruction satisfies the requirement that its SNR is superior to 5 in the spectral range from 570 to 580 nm. The SNR value estimated for each reflectance spectra is then between 5 and 107, with an average value of 38 and a standard deviation of 22.

Fig. 5

(a) Representative in vivo diffuse reflectance spectra and (b) optical sinogram associated with an interstitial measurement in swine brain.

Figure 6 shows a 3-D representation of the blood vessel distribution detected in swine brain using CT and diffuse optical imaging. For comparison purposes, each of the nine diffuse optical 2-D images [Fig. 6(a)] are juxtaposed with their corresponding 2-D CT slice from the full 3-D volume to assess the potential of the optical needle system to detect vessels. Importantly, all diffuse reflectance spectra are inspected and they show characteristic features of Hb, confirming that the optical contrast is not provided by the x-ray contrast agent itself, which had cleared the vessels’ network at the moment when the diffuse reflectance images are acquired. The Hb features themselves are the high optical absorption in the near 570 to 580 nm region and the low absorption elsewhere in the spectral range of the imaging system. An experienced radiologist (GS) analyzed the CT images to discriminate vessels from image noise and was able to confirm that the contrast observed in Fig. 6(b) is associated with blood vessels. For the slices at depth 3, 6, 15, 18, 21, and 23 mm, both the diffuse optical image and the CT slice are consistent with an absence of blood vessels. There is also a match for depths 0 and 12 mm where contrast is observed in both the diffuse optical and the CT images. However, the slice at depth 9 mm can be considered as a false negative for diffuse optical image as the CT scan image shows blood vessel contrast not detected with the optical imaging system.

Fig. 6

Comparison of spatially coincident images: (a) diffuse optical reconstructions and (b) CT scan ($x$-axis and $y$-axis units are in millimeters).