2.1.

Intraoperative MRI

ioMRI-guided intracranial surgery was first applied in late 1990s.^21,22 High-field MRI with an integrated neuronavigation system is one of the most advanced nonoptical intraoperative imaging methods.²³ It allows for objective control of the extent of resection and for immediate update of the neuronavigation data.^24,25 Anatomical and functional information enable the surgeon to decide whether it is necessary and safe to proceed with tumor resection. IoMRI allows more precise delineation of the tumor margin and detects hemorrhage.^26,27 It also provides an objective measure of the needed extent of surgery and allows more radical, yet safe, tumor removal, regardless of tumor type.²⁸ Its advantages over ultrasound or CT-scan-guided resections include lack of ionizing radiation, superior multiplanar imaging, and user independence.²⁹ IoMRI is rapidly becoming the procedure of choice for many neurosurgeons because of advantages of real-time high-quality images and ability to account for brain shift. However, this technique contains many disadvantages as well, including cost and time constraints. IoMRI systems are usually too expensive for most neurosurgical departments, involve a time-consuming image acquisition process, and require expensive nonmagnetic surgical tools.¹³ Furthermore, using these techniques, it is not possible to track surgical instruments as they pass through brain tissue to cysts or tumors.³⁰ Finally, after resection of the tumor and the resulting brain shift, the interpretation of the MRI data can still be difficult. The use of ioMRI for neurosurgical procedures also involves a gradual learning curve to acquire sufficient expertise,²⁸ and if intraoperative imaging suggests there is residual tumor present, then surgeons must apply their own clinical experience and expertise to each specific situation.

2.2.

Intraoperative Ultrasound

ioUS technology is useful because it is a real-time imaging system that remains mostly unaffected by brain shift. It is also very easy to handle and inexpensive, two important advantages over ioMRI.^31–33 The two-dimensional (2-D) US image acquisition is significantly faster than ioMRI and scans can be performed multiple times during surgery, without major interruption of the workflow.²⁴ US imaging can be also integrated with neuronavigation systems.³⁴ Today the technology and image quality of ultrasound have been improved significantly, providing the surgeon with a useful way to identify tumors and make a preliminary classification of low-grade or high-grade glioma.³⁵ Its sensitivity and specificity in detecting larger or deeply located tumor remnants are comparable to those of high-field ioMRI.³⁶ Some studies have even reported the ability of ultrasound to differentiate edema from solid tumor, a task that is difficult for both CT and MR imaging.³⁷ However, two shortcomings are that ioUS provides lower sensitivity and specificity in the detection of small or superficial tumor remnants.²⁴ Furthermore, the image acquisition and interpretation are subjective and require adequate training and experience.²⁴ Other confounding factors that may affect the interpretation of the ioUS image are the presence of dropouts from blood, cottonoids, or proteinaceous fluid, which can all reduce the ultrasound image quality, or the contrast of the disrupted BBB.³¹

3.1.

Thermal Imaging

Intraoperative thermal imaging (ioTI) is a neuroimaging technique that has the potential for locating the margins of brain tumors. The real-time anatomical and pathophysiological information provided by ioTI may contribute to safely improving the extent of tumor resection.⁴⁴ The presence of a tumor often alters the homeostasis of the surrounding tissue and ioTI measurements are sensitive to this change. Changes in homeostasis that affect temperature can be due to low density in tumor microvessels,^45,46 edema in the surrounding tissues due to increases in the volume of white matter caused by tumor growth,^45,46 lower metabolism in the cortex overlying intracranial tumors, due to the “disconnection effect” caused by edema,^45,46 or tumor necrosis, typically caused by reduction of tumor vasculature and metabolic activity.

Although not as common as ioMRI, ioTI has been used in several experimental models and case reports for brain tumor detection. In both animal and human models, glial-derived brain tumors display a consistently lower temperature than the surrounding normal tissue.^45,46 In patients, Gorbach et al.⁴⁶ showed that these tumors generally have a decrease of 0.5 to 2.0°C compared to normal surrounding tissue. Kateb et al. showed a case study of a 76-year-old female patient with metastatic intracortical melanoma. The data revealed a clear demarcation of tumor with significant temperature differences, up to 3.3°C, between the tumor core (36.4°C) and the surrounding normal tissue (33.1°C) when intraoperatively measured by a ThermaCAM P60 FLIR system (wavelength sensitivity: 7.5 to $13\mathrm{\mu }\mathrm{m}$).⁴⁴

Thermal imaging provides a surface recording of a tissue’s heat profile with low penetration, so the use of ioTI requires clear exposure of the lesion. Neural activity, local metabolism, blood flow, and thermal conductivity are all factors that contribute to the thermal profile of a tissue. ioTI is a versatile and useful intraoperative tool that is cost-efficient and noninvasive. Limitations of this technology include a lack of software that can autodelineate the tumor margin, a need for cameras with higher spatial resolution, and limited tissue penetration depth ($<0.5\mathrm{mm}$).⁴⁴ Ultimately, infrared thermography could become a vital addition to high-resolution neurosurgical microscopes that can accurately distinguish between normal and cancerous tissue using diverse multimode technologies.

3.2.

Raman Spectroscopy

Raman spectroscopy is a nondestructive spectroscopic technique, based on the vibrational properties of the constituent molecules, that provides molecular information about the tissue under examination.⁴⁷ The Raman signal results from tissue molecules being excited by a small amount of incident light at a specific wavelength. The remitted light has some of its photons shifted to different wavelengths as by the addition or subtraction of vibrational energy from some of the tissue intramolecular bonds.⁴⁸ Contrast is achieved when the tissue molecular constituents differ enough that the Raman signals from two tissues have different wavelength distributions. Raman spectroscopy may be an ideal tool to intraoperatively delineate gray matter from white matter because it is minimally influenced by water content and can be performed in near-real time with sterile optical probes that a surgeon can apply to tissue in the operative field.⁴⁷

Raman spectroscopy imaging techniques have been used to identify histologic features in brain tissue frozen sections of rat glioma. The rat glioma model has similar morphology to glioblastoma multiforme.⁴⁹ This study showed that Raman spectral features could be used to differentiate tumor, edema, necrotic, and healthy brain tissue. Specific spectral bands associated with proteins and lipids are the main Raman spectral features used for differentiation.

Recently, an intraoperative Raman-based spectroscopy technique detected invasive brain cancer in $<1\mathrm{s}$ during surgery and detected grade 2 to grade 4 gliomas in vivo.^50,51 It distinguished brain tissue invaded by cancer from normal brain tissue with a sensitivity of 93% and specificity of 91% (17 patients). The point spectroscopy method used a handheld fiber probe illuminating a 0.5-mm-diameter tissue area. The probe sampled depth up to $\sim 1\mathrm{mm}$ with acquisition time for each measurement requiring about 0.2 s. These Raman signatures are based on lipid bands (700 to $1142{\mathrm{cm}}^{-1}$), nucleic acid bands (1540 to $1645{\mathrm{cm}}^{-1}$), and the phenylalanine band in proteins ($1005{\mathrm{cm}}^{-1}$). Machine learning algorithms created a classifier to differentiate between normal and tumor tissue types.

As shown in Fig. 1(a), the probe contained optical fibers connected to a NIR spectrum-stabilized laser ($\lambda =785\mathrm{nm}$) for inject energy and a fast CCD spectrometer for detection of the returning Raman signal. During each resection, the Raman signal was measured at several points in the surgical cavity [Fig. 1(b)].

Fig. 1

Handheld contact fiber optic probe for Raman spectroscopy. (a) Experimental setup diagram. (b) A simple molecular vibrational mode is conceptually depicted (individual atoms in blue and green) interacting with the laser light (in red) to produce Raman scattering (in purple).

Figure 2(a) shows a 3-D preoperative MRI of a patient with a grade 2 glioma, with dots identifying the Raman measurement locations (blue for normal and yellow for tumor). For three of the locations, the 2-D MRI slices are provided with corresponding pathology images, as well as the Raman spectra [Fig. 2(b)] for dense cancer, invasive cancer, and normal brain. Sample Raman spectra, preoperative MRI, and pathology images are also provided for a patient with a grade 3 glioma [Figs. 2(a) and 2(b)] and a patient with a grade 4 glioma [Fig. 2(C)].

Fig. 2

Raman spectroscopy measurements colocated on preoperative MRI-grade 2 and 4 gliomas. (a) 3-D rendering of a preoperative T2-weighted MRI overlaid with the segmentation of a grade 2 astrocytoma in red. Samples for each tissue type are indicated by P1, P2, and P3. (b) Pathology images for regions P1 to P3 in (a). P1, P2, and P3 are dense cancer, invasive cancer, and normal brain, respectively. The acquired Raman spectra are shown below. (c) In a different patient with grade 4 GBM, locations for dense cancer, invasive cancer, and normal brain are shown on a T1-weighted MRI with corresponding histopathology.

Cancer cells showed some differences in various bands including (1) the lipid bands at 700 and $1142{\mathrm{cm}}^{-1}$ corresponding to cholesterol and phospholipids (26), (2) an increase from 1540 to $1645{\mathrm{cm}}^{-1}$, corresponding to a higher nucleic acid content, and (3) an increase at the $1005{\mathrm{cm}}^{-1}$ band, corresponding to phenylalanine breathing mode (26).

The new field of Raman-based optical biopsy for surgical guidance provides minimal disruption to workflow. Current development of more sensitive detectors and fast electronics shows the promise to overcome the low signal-to-noise ratio of the Raman signal, leading to Raman-based imaging. Future work will continue to explore Raman signal uses for brain tissue discrimination and surgical guidance in larger clinical trials.

3.3.

Optical Coherence Tomography

OCT is a noninvasive, label-free, and cost-effective technique capable of imaging tissues in 3-D in near real time.⁵² OCT has spatial resolution in the micrometer range for depths up to a few millimeters. Clinical applications of OCT have been mainly in ophthalmology, however, over the last 20 years it has been extended to cancer detection in a variety of tissues.^53,54 One of the first studies on ex vivo brain tissue by Bizheva et al.⁵⁵ was reported in which the OCT signal was used to distinguish between healthy brain and multiple types of brain tumors. This study claims the source of OCT signal contrast comes from differences in cell nuclei, cysts, microvasculature, and microcalcifications in brain tissue.

Recently in 2015, structural and functional information provided by OCT has been explored to study brain diseases and tumors in vivo.⁵⁶ In a study by Kut et al., OCT was used on ex vivo human tissue samples of 32 patients with grade 2 to 4 gliomas. They used attenuation-based thresholding to distinguish between normal and tumor tissue types. The threshold values were obtained by a researcher blinded to tissue type and grade. Specificities and sensitivities of 100% and 92% (for high-grade glioma) and 80% and 100% (for low-grade glioma) were reported.

In the same article, they have showed the extension of this work to an in vivo study with a cohort of nine subjects. They observed distortions in tissue microstructure and light attenuation profiles in the tumor region that were confirmed by biopsy and histopathological examination. The OCT system can measure an 8- to $16\text{-}{\mathrm{mm}}^3$ tissue block in 1.2 to 2.4 s, making it a useful tool with minimal disruption during neurosurgery.

Figure 3(a) shows the swept-source OCT (SS-OCT) system and the handheld imaging probe that were used in the study. The OCT probe design offers a working distance of 25 mm and a confocal parameter of 0.3 mm [see Fig. 3(b)]. To remove the influence of depth-dependent effects on the beam profiles, they normalized the OCT signal from the tissue with the OCT signal from a phantom with known attenuation properties. The averaged OCT signal values of the biological tissue and phantom at different depths are used to calculate optical attenuation that provided the threshold parameter to differentiate between normal and cancerous tissues. An optimal attenuation threshold was established for identification of high-grade and low-grade human brain cancer. Finally, to facilitate potential intraoperative use of this technology, OCT images were displayed in 3-D and color-coded optical property maps showing tissue attenuation properties [Fig. 3(c)].

Fig. 3

(a) A SS-OCT system for intraoperative neurosurgery procedure. (b) A 2-D scanning handheld imaging probe. (c) Characteristic curves were determined using brain tissues in the training data set. A 3-D volumetric reconstruction map from the OCT images with an overlaid color-coded attenuation map distinguishing cancer and noncancer regions.

3.4.

Fluorescence-Based Technologies

3.4.1.

Endogenous fluorescence in tissue

There are several endogenous fluorophores that exhibit potential for analyzing tumor environments. These include amino acids such as tyrosine and tryptophan, structural proteins such as elastin and collagen, and enzyme cofactors [e.g., reduced nicotinamide adenine dinucleotide (NADH) and flavin-adenine dinucleotide (FAD)]. The altered metabolic state seen in brain tumors causes changes in these molecules’ fluorescence signatures, which can be utilized for tissue characterization. By example, NADH concentration increases in tissues with higher metabolic activity. NADH displays fluorescence emission in the spectral range of 400 to 700 nm when exposed to UV radiation, whereas its oxidized form (NAD+) is nonfluorescent. In addition, the lifetime of NADH is either short or long, depending on the state of its protein binding sites. Changes in the ratio of free-to-bound NADH in tumor tissue results in fluctuations of the average lifetime of NADH.⁵⁷ Thus, both the intensity and lifetime of NADH fluorescence are considered to be potential indicators for monitoring the tumor microenvironment. Most endogenous fluorophores are excited by UV-VIS light (280 to 450 nm) and emit fluorescence in the range of 370 to 700 nm, as shown in Fig. 4.^58,59 In general, tumor necrotic regions have lower fluorescence emission compared to normal tissue, under UV excitation at 355 nm, likely due to lower tissue metabolism. One experimental approach demonstrated that the ratio of NADH:FAD fluorescence emission may play a significant factor in identifying brain tumors.⁶⁰

Fig. 4

Natural fluorophores of biological tissue.

3.4.3.

Time-domain fluorescence measurement methods

Time-resolved fluorescence determines the dynamics and intensity of fluorescence decay (lifetime). Fluorescence lifetime is the average time when a fluorophore remains in the excited state after being excited by the light source. It could be a more effective tool for optical biopsy because not only does the contrast from different fluorescence lifetimes result from changes of cellular or tissue environment, but also these changes are independent of fluorescent signal intensity, irregular brain surfaces, tissue illumination, blood, and fluorophore quantum yield.^61,66 Time-resolved fluorescence measurement can also differentiate the fluorescence components of overlapping emission spectra, providing the ability to detect variations in pH and temperature.^59,67 However, time-resolved fluorescence measurements (either spectroscopic or imaging) require more complex equipment, including more sophisticated lasers, high-speed detectors, and electronics that function in the picosecond range. A short-pulsed laser is widely used as the excitation light source for time-domain measurement, whereas a modulated light source is used for frequency-domain measurement. Single-point spectroscopy^68–70 and multispectral/hyperspectral imaging systems^71–73 have both been investigated for their potential use in tissue diagnosis. In the single-point method, a fiber optic probe delivers laser light and collects fluorescence signal, making time-resolved fluorescence spectroscopy (TRFS) much more compatible with the clinical setting. In imaging techniques, a single fiber or fiber bundles are used to collect the image in wide-field-based fluorescence-lifetime imaging microscopy (FLIM),^74,75 whereas high-speed scanning modules (e.g., galvanometer mirrors) are used in scanning-based FLIM.⁷²

Time-resolved fluorescence imaging

In this method, the fluorophore is excited by a short-pulse laser, whereas emission is detected at different times along the decay curve by a fixed-gating time imaging detector (e.g., 200 ps), which is then used to reconstruct the decay curve in the time domain. With the use of a time-gated intensified charge-coupled device (ICCD), this method is suitable for obtaining an image over a large spatial area. Sun et al. reported the application of fluorescence lifetime imaging to glioblastoma detection during surgery. They reported weaker fluorescence emission intensity and longer lifetime at 460 nm for GBM compared to normal cortex for a limited number of patients ($N=3$).⁷⁵ They evaluated the FLIM measurements with MRI imaging as well as pathology assessment. However, both the temporal resolution and the sensitivity were relatively low.

Time-resolved fluorescence spectroscopy

Time-resolved fluorescence spectroscopy (TR-FS) is a technique that uses ultrashort laser pulses to excite a tissue sample and record the corresponding fluorescence intensity decay in multiple spectral bands.⁷⁶ The fluorescence decay is a robust measure of tissue characteristics and is not affected by the fluorescence intensity fluctuations commonly found in intraoperative data. This gives TR-FS an advantage over other optical techniques that can be susceptible to blood absorption, measurement distance from probes, and surgical microscope background light.

Butte et al.⁷⁷ reported using TR-FS to detect fluorescence lifetime differences between normal brain tissue and meningioma in 22 patients with a sensitivity/specificity ratio of 89/100. Their statistical analysis was based on one-way analysis of variance (ANOVA) applied to spectroscopic parameters, such as the ratio between the various fluorescence intensities, and time-dependent parameters observed near the peak at 390 nm. In 2011, they reported on the potential of applying TR-LIFS for intraoperative diagnosis of gliomas.⁷⁸ Normal cortex ($N=35$) and white matter measurements ($N=12$) showed longer-lasting fluorescence emission at 390 nm compared with 460 nm, whereas the 390-nm emission peak was absent in low-grade gliomas and reduced in high-grade gliomas. The system required a relatively long data acquisition time ($\sim 30\mathrm{s}$) to capture each tissue measurement, which would need to be shortened substantially for effective integration in clinical workflows.⁷⁹

Vasefi et al.⁸⁰ recently reported on an improved TR-FS system with near real-time data acquisition and classification. In this study, tissue differentiation on larger tissue samples measured in vivo NC ($n=10$), WM ($n=12$), and GBM ($n=13$) with nine patients is described. The TR-FS system (shown in Fig. 5) used a 355-nm laser pulse (pulse width: 350 ps, pulse rate: 1 kHz) to excite tissue fluorescence and an optical demultiplexer to separate the emission photons into six wavelength bands from 360 to 700 nm. Since multiple detectors would be costly and require individual calibration, each individual color band is passed through a fiber delay unit. The fiber delay units have fiber optic coils of varying lengths. The longer the fiber, the more delay occurs before the signal reaches the single system detector. This produces a train of signals containing different color band information reaching the detector at different times.

Fig. 5

TRFS system diagram.

The fluorescence decay function was calculated by deconvolving the fluorescence measurement by instrument response function. For each TR-FS measurement, six decay curves were measured corresponding to each color band. Since the fluorescence decay may contain multiple intrinsic fluorophores in tissue, the decay function cannot be presented by only average lifetime scale. An intrinsic fluorophore may have a monoexponential, biexponential, or multiexponential decay function. Furthermore, there may be $>1$ intrinsic fluorophore that can further complicate the decay function analysis.

This method transforms the decay functions at different wavelengths into a 2-D spectro-lifetime matrix (SLM) with $m\times n$ dimensions, where $m$ is the number of spectral bands used in the measurements and $n$ is the number of decay points used. Figure 6 shows the three lifetime functions in six color bands corresponding to slow (${\tau }_{[0.2]}$), middle (${\tau }_{[0.4]}$), and fast (${\tau }_{[0.6]}$) decay components of fluorescence decay measured in human brain tissue. The lifetime values are separated by crossing normalized fluorescence impulse response function at 0.2, 0.4, and 0.6 intensity levels.

Fig. 6

(a) Spectral signatures of fast, (b) normal, and (c) slow lifetime for brain tissue measured by TRFS system.

Figure 7 shows classification results distinguishing NC, WM, and GBM measurements acquired in vivo. Linear discriminant analysis with a three-group subclassifier was implemented. The first subclassifier identified whether the TR-FS measurement is NC or “not NC.” The second or third subclassifiers identified whether the measurement is WM or “not WM,” or GBM or “not GBM.” Figure 7 shows the ability of each subclassifier to discriminate, for a training group, between NC ($N=10$, blue), WM ($N=12$, green), and GBM ($N=13$, red) tissue measurements in nine patients. This shows that SLM data can be used to effectively discriminate all three tissue types, provided the classifier is not over trained. The classifier algorithm was tested on eight NC measurements from a different patient with meningioma (Fig. 8).

Fig. 7

TRFS classification output for differentiating normal cortex, white matter, and GBM tissue types.

Fig. 8

Meningioma case study with normal cortex measurements ($N=8$).

The near real-time speed of the system makes it much more usable during a clinical trial, with minimal interruption during a surgical procedure. Validation of the TR-FS system in larger trials is still required before technology adoption.