Applications of Ultrasound in the Resection of Brain Tumors

Abstract

Neurosurgery makes use of pre-operative imaging to visualize pathology, inform surgical planning, and evaluate the safety of selected approaches. The utility of pre-operative imaging for neuronavigation, however, is diminished by the well characterized phenomenon of brain shift, in which the brain deforms intraoperatively as a result of craniotomy, swelling, gravity, tumor resection, cerebrospinal fluid (CSF) drainage, and many other factors. As such, there is a need for updated intraoperative information that accurately reflects intraoperative conditions. Since 1982, intraoperative ultrasound has allowed neurosurgeons to craft and update operative plans without ionizing radiation exposure or major workflow interruption. Continued evolution of ultrasound technology since its introduction has resulted in superior imaging quality, smaller probes, and more seamless integration with neuronavigation systems. Furthermore, the introduction of related imaging modalities, such as 3-dimensional ultrasound, contrast-enhanced ultrasound, high-frequency ultrasound, and ultrasound elastography have dramatically expanded the options available to the neurosurgeon intraoperatively. In the context of these advances, we review the current state, potential, and challenges of intraoperative ultrasound for brain tumor resection. We begin by evaluating these ultrasound technologies and their relative advantages and disadvantages. We then review three specific applications of these ultrasound technologies to brain tumor resection: (1) intraoperative navigation, (2) assessment of extent of resection, and (3) brain shift monitoring and compensation. We conclude by identifying opportunities for future directions in the development of ultrasound technologies.

Introduction

Ultrasound (US) was first applied to adult neurosurgery in 1982, at which time the advent of 2-dimensional B-mode imaging (2D US) allowed real-time visualization of neural anatomy and pathology during surgical interventions.¹ Since then, intraoperative ultrasound has allowed surgeons to craft and update operative plans with without ionizing radiation exposure or major workflow interruption. Continued evolution of ultrasound technology has yielded superior image quality, smaller probes, and more seamless integration with neuronavigation systems. Furthermore, the introduction of related imaging modalities, such as 3-dimensional ultrasound (3D US), contrast-enhanced ultrasound (ceUS), high-frequency ultrasound (hfUS), and ultrasound elastography promise additional neurosurgical benefits (Table 1). In the context of these advances, we review the current state, potential, and challenges of intraoperative ultrasound for brain tumor resection.

Table 1.

Advantages and limitations of ultrasound (US) imaging modalities.

Technology	Advantages	Limitations
2D US	Speed of image acquisition Simplicity Balance of resolution and depth of penetration Low cost	Difficulty of integration with 3D imaging High inter-operator variability Similarity of appearance of tumor and chronic edema
3D US (asynchronous)	Volumetric acquisition Possible with standard 2D US transducer Flexible field of view	Slow acquisition and construction of image Significant potential for motion artifact Very high inter-operator variability
3D US (mechanical translation)	Volumetric acquisition Regularity of acquisition minimizes artifacts Faster than asynchronous 3D US	Restricted field of view Slower than 2D US
3D US (phased- array)	Volumetric acquisition Speed of acquisition is comparable to that of 2D US	Diminished image quality
High-frequency US	High resolution More reliable visualization of tumor margins, especially in the context of peritumoral edema	Difficulty of integration with 3D imaging Poor depth of penetration
Power Doppler US	Visualization of tumor vascularity	Poor resolution High noise High potential for inter- operator variability
Contrast-enhanced US	Visualization of tumor vascularity	Field of view is constant during contrast injection High potential for inter- operator variability
US Elastography	Visualization of tumor elastic properties	Unclear correlation with pathology Induced strain risks trauma to brain tissue High noise

Technical Background

2-Dimensional Ultrasound

Ultrasound systems use sound waves at frequencies higher than those detectable by human hearing, or approximately greater than 20 kHz. Typical medical ultrasound systems utilize an array of piezoelectric transducers that emit pulses at a frequency of 1–20 MHz. These pulses are absorbed, scattered, or reflected as a function of the intrinsic acoustic properties of tissue in their paths. These same transducers detect wave echoes and allow the production of a 2-dimensional image based on the time delays from the transmitted pulses to the received echoes. Depending on the orientation of the transducer array, the resulting image may be rectangular, fan-shaped, or pie-shaped. In general, high-frequency transducers produce higher resolution images but provide less tissue penetration due to scattering and absorption. In many neurosurgical applications, US transducers operate at 5–10 MHz and provide pixel resolutions of 500–1100 μm at depths of 2–8 cm from the transducer.^2–6 Newer, high-frequency transducers, by comparison, operate at frequencies up to 25 MHz and can provide maximum pixel resolutions of 100–600 μm but only at depths of 2–4 cm from the transducer.^3,4,7 The optimal choice of transducer type and acquisition frequency will depend on the location and sonographic properties of the tumor, the size of the craniotomy, surrounding anatomy, surgeon preference, and other factors.

The brightness of structures in an US image reflects the amplitude of the reflected signal. Acoustically homogeneous tissue generates little signal while structures with high gradients of acoustic impedance generate strong echoes and can obscure structures further from the probe. In the normal brain, a variety of structures with strong acoustic gradients, including sulci, falx cerebri, choroid plexus, and vessel walls, appear hyperechoic on US imaging. Acoustically homogeneous regions such as ventricles, cysts, and other spaces filled by cerebrospinal fluid (CSF) are hypoechoic. Brain parenchyma is relatively uniform, with grey matter appearing slightly hyperechoic relative to white matter. Tumors are frequently hyperechoic on ultrasound imaging as a result of their relatively high mass density. Acute edema appears hypoechoic, allowing it to be distinguished from mass lesions. Chronic edema, however, is considerably more unpredictable in appearance, creating a potential confounder during intraoperative ultrasound guidance of resection.⁸

Equipment and Technique

Ultrasound has been utilized for intraoperative imaging in a variety of settings including liver,⁹ pancreatic,¹⁰ breast,¹¹ and laparoscopic surgery.¹² In neurosurgery, US has been used at various points during tumor resection for localization and characterization of tumor, for operative planning, and for assessment of extent of resection. The application of US to brain tumor resection, however, is uniquely constrained by the size of craniotomy, which can be very small relative to the footprint of many US probes, and the creation of a resection cavity, which can eliminate a useful contact surface between the probe and the brain. Accordingly, in the literature, both surgical and non-surgical (abdominal, endovaginal, cardiologic, and pediatric) US probes have been applied in brain tumor resections.

Guidelines regarding the use of intraoperative ultrasound in hepatic surgery recommend that intraoperative probe and its cable be either sterilized or, if not possible, covered with sterile sheath containing sterile coupling gel.¹³ US Probes can be sterilized using one of three major approaches: ethylene oxide gas, hydrogen peroxide vapor-gas plasma, or glutaraldehyde immersion.¹⁰ The use of ethylene oxide sterilization, one of the most common methods, is limited by potential damage to the probe as a result of high temperature exposure and by relatively long turnaround times (16–24 hours).¹⁴ Glutaraldehyde immersion has faster turnaround times (4 hours) and no heat exposure; however, its use is waning due to its irritant and allergenic properties when in contact with skin, eyes, and respiratory tract.^14,15 Hydrogen peroxide vapor-gas plasma offers quick turnaround (2–3 hours) without the risk of contact reaction.^10,14 Nearly all US probes used in the literature can be sterilized with glutaraldehyde immersion regardless of their original design for surgical or non-surgical applications (Table 2). Comparatively few probes could be safely sterilized by hydrogen peroxide vapor-gas plasma.

Table 2.

Selected ultrasound systems and probes: Selected ultrasound systems and probes reported from the literature with manufacturer-supported methods of sterilization

Manufacturer	System	Probes	Type	Glutaralde hyde Sterilization	Ethylene Oxide Sterilization	Hydrogen Peroxide Sterilizatio
Aloka	alpha-10	UST-9133	2D	yes	yes	yes
BK	3500	8862	2D	yes	no	yes
Esaote	MyLab	LA 332	2D	yes	no	no
GE	LOGIQ E9	9 L-D	2D	yes	no	no
GE	LOGIQ E9	IC5-9D	2D	yes	no	no
Philips	iu-22	x7-2	3D Phased Array	yes	no	no
Philips	iu-22	L15-7io	HF 2D	yes	no	no
Philips	iu-22	X3-1	3D Phased Array	yes	no	no
Philips	EPIQ	c10-3 V	2D	yes	no	no
Siemens	Acuson	8C4	2D	yes	no	no
Siemens Sonoline	Sienna	C8-5	2D	yes	no	no
Toshiba	Aplio XG	PLT704SBT	HF 2D	yes	no	yes
Toshiba	Aplio XG	PVT745BTV	2D	yes	yes	yes
Toshiba	Aplio XG	PST65AT	3D	yes	yes	yes

At our institution, we use a sterilized probe without a sterile cover. If used, the sterile cover should be fit tightly to the probe to minimize artifacts.^8,10 The sterilized or covered probe can be applied to dura or to exposed brain, with or without sterile gel or saline for acoustic coupling, in order to visualize the anatomy and the tumor prior to resection.¹⁶ During imaging, the surgeon must be careful to not apply too much pressure, which would cause deformation and limit the utility of the image⁸ Once a resection cavity has been established, it may be possible to insert a sufficiently small US probe into the cavity to visualize any remaining tumor; however, in many cases, the cavity should be filled with sterile saline to acoustically couple a relatively large transducer, which is held in contact with the saline at the level of the cortical surface.¹⁷ Positioning the patient such that the resection surface is horizontal with respect to gravity helps to minimize air trapping within the cavity.⁵ Other approaches to intraoperative imaging, including making a larger craniotomy and making a separate craniotomy, have also been described.⁵

Neuronavigation and Tracking

Integration of ultrasound with neuronavigation technology has enabled side-by-side visualization of ultrasound with corresponding preoperative imaging.¹⁸ Standard tracking systems involve the use of optical or magnetic trackers as used in neuronavigation systems physically mounted to the ultrasound probe (Figure 1). In spite of the promise of MRI/US fusion to improve brain tumor resection, adoption and technical progress have been limited. Currently available systems for neurosurgery that support tracked 2D US suffer from high technical complexity and artifacts, which limit image interpretation. Research to improve MRI/US fused neuronavigation is hampered by proprietary data formats and interfaces provided by available equipment. While the Digital Imaging and Communications in Medicine (DICOM) standard provides a vendor-neutral encoding for volumetric US datasets,¹⁹ we are not aware of any commercial neurosurgical US systems supporting the standard. As a result, researchers developing next generation neuronavigation systems have resorted to development of custom interfaces, in some cases, reverse-engineering proprietary data formats and mapping them to open standards, in order to perform experiments in this field.^20,21

Figure 1. Tracked 2D ultrasound with superimposed color Doppler imaging during tumor resection.

Legend: (A) Intraoperative 2D tracked ultrasound image acquisition in the operating room. Fiducials for intraoperative optical neuronavigation are visible rigidly attached to the ultrasound probe. (B) 2D ultrasound image with superimposed color Doppler imaging.

Artifacts and Limitations

Successful ultrasound image acquisition and interpretation can be a challenging for surgeons who lack experience with the technology. Training programs in ultrasonography, practice on phantom or animal models, and consultation with expert sonographers is recommended in order to refine surgical technique and avoid US artifacts. Artifacts, defined as any part of an US image that does not accurately represent the anatomy of the structure being visualized, arise for a variety of reasons. As previously noted, tissues with high signal-attenuation, such as bone, can create a shadow that diminishes signal of any tissue behind it. Acoustical interfaces at excision boundaries also pose significant interpretation challenges. Specifically, the attenuation coefficient of saline, which is often used to fill the resection cavity, is lower than that of neural tissue, producing a brightness artifact at the brain/saline interface, and potentially obscuring the identification of residual tumor. Coagulated blood or hemostatic materials located along the walls of the resection cavity can also produce a brightness artifact. Alternative fluids with acoustic properties that mimic brain tissue are under investigation to improve the quality of ultrasound imaging during tumor resection.^17,22 Furthermore, variation in the speed of sound in tissues can cause geometric distortion of the 2D image compared to the actual anatomy. While brain tissue itself is relatively homogeneous, the presence of saline, which conducts sound at a slower speed than brain, can produce an error of approximately 1.6 mm at a depth of 10 cm from the transducer.¹⁷ Even the temperature of the saline can change the speed of sound and thereby influence image formation.

Overall, the low cost, ease of use, and real-time feedback of 2D US have made it a useful adjunct during tumor resection. 2D US does, however, have some notable limitations. The resolution of US is not uniform in every direction and depends on depth of focus and other parameters; as such, the location of an object within a frame influences the resolution at which it can be imaged.²³ Because the appearance of tissue depends on the depth and angle of incident US waves, it can be challenging both to interpret US images and to compare them to images from other modalities, such as MR and CT. Furthermore, multimodality image interpretation can be challenging because a 2D US image is formed in the plane perpendicular to the US transducer face and this plane typically does not match the standard axial, sagittal, or coronal sections with which surgeons are familiar. Neuronavigation systems can help address this issue by tracking the ultrasound acquisition plane and displaying preoperative images reformatted to match the US acquisition plane. Nonetheless, inter-operator variability in the interpretation of multimodality images can be significant.

3-Dimensional Ultrasound

3D ultrasound, which was initially developed for prenatal applications, overcomes some of the limitations of 2D US by producing a volumetric image that can be viewed in any arbitrary plane (Figure 2).²⁴ 3D US volumes can be produced in three major ways: freehand acquisition, mechanical sweeping, and phased array transducer.

Figure 2. Tracked 3D ultrasound (US) imaging and MRI during glioma resection at the Advanced Multimodality Image Guided Operating (AMIGO) Suite at Brigham and Women’s Hospital.

Legend: (A) preoperative T2-Fluid-attenuated inversion recovery (FLAIR) MRI, (B) Pre-durotomy intraoperative 3D US overlaid on preoperative T2-FLAIR MRI, (C) Pre-durotomy intraoperative 3D US, (D) intraoperative T2-FLAIR MRI following initial tumor resection, (E) intraoperative 3D US following initial tumor resection overlaid on intraoperative T2-FLAIR MRI, and (F) intraoperative 3D US following initial tumor resection.

The earliest and simplest method is asynchronous freehand 3D US, in which a volume is reconstructed from a series of 2D US images acquired as the probe is swept across the target anatomy.²⁵ The volumes are asynchronous in the sense that they are only created after the surgeon has acquired the set of images and, therefore, not in real time. Tracked neuronavigation is often necessary to align the 2D frames in space for reconstruction; however, variation in pressure applied to the brain during freehand acquisition can still produce artifacts. A more recent innovation in 3D US is mechanical volume sweeping.²⁶ In such systems, a small, mechanical device moves the probe head evenly along a predefined trajectory. As compared to the freehand approach, mechanical sweeping produces higher resolution images with shorter reconstruction times that are limited only by the time taken by the motorized device to sweep the field; however, the fields of view generated with such systems are restricted to the predefined sweep trajectory.²⁴ 3D phased array transducers, in which a 2D array of transducers are used to produce a pyramidal image volume, are becoming available for 3D US image generation.^27,28 Unlike the previous two methods, the phased array approach produces 3D images directly, thereby minimizing motion artifacts and supporting real-time visualization; however, currently, natively 3D systems to not provide the same in-plane resolution as comparable 2D US systems.²⁴

In general, 3D US methods can be used for the same purposes in tumor resection as 2D US methods. The artifacts described above with respect to 2D US apply equally or more to 3D US images based on 2D US acquisitions. In particular, because the resolution of 2D US frames is non-uniform, arbitrarily reslicing 3D image volumes along axes that differ from the original acquisition planes results in diminished or uneven image resolution.²³ Furthermore, because the appearance of a given object depends on the properties of intervening tissue, determining the correspondence of 3D reconstructions of anatomy visualized from two different angles can be particularly challenging. Reconstruction quality is subject to inter-operator variability and highly sensitive to motion artifacts. Accurate interpretation of these images requires mindfulness of the potential sources for reconstruction errors inherent in the acquisition method.

High-frequency Ultrasound

High-frequency linear array US transducers operate at frequencies up to 25 MHz to provide higher resolution images and, potentially, more reliable demarcation of tumor margins compared to conventional US even in settings of peritumoral edema and prior radiotherapy.^2–4,7,29 While applications of high-frequency US have been limited by poor depth of view, newer transducers are small enough to be inserted into resection cavities for more direct visualization of the target.³ Given the successful utilization of high-frequency US in breast cancer resection,³⁰ the potential value of more accurate visualization in brain tumor resection is significant. However, the ability to visualize only small parts of the resection cavity and surrounding tissue at a time presents a significant impediment to the intraoperative use of this technology compared to standard 2D US.

Doppler Ultrasonography

In brain tumor resection, Doppler ultrasonography can be used to assess tumor vascularity and to plan the operative approach. Doppler ultrasonography utilizes the Doppler effect, which is an observed frequency shift when a US wave is reflected back to the transducer from moving particles, to determine the direction and relative velocity of fluid along the axis of the probe. There are subtypes of Doppler imaging that are based on this general principle. Color Doppler imaging relies on the magnitude of measured Doppler shift and, in a selected portion of the US frame, shows with color the direction of flow, either toward or away from the probe, overlaid on the 2D US image (Figure 1). As such, color Doppler is very angle dependent: at any point where flow is perpendicular to the US waves, no Doppler shift and therefore, no flow, will be observed.³¹ Therefore, a change in angle and a change in velocity may appear similarly on color Doppler. Furthermore, color Doppler suffers from aliasing, an artifact in which areas of flow are represented with incorrect magnitude or direction as a result of transducer pulse rate limitations.³¹ Finally, color Doppler imaging is highly susceptible to noise, which may overwhelm the flow signal.³²

Power Doppler, which relies on the power of the Doppler shift signal instead of the magnitude of the shift, was developed as an alternative to color Doppler imaging.^31,32 As compared to color Doppler, power Doppler has less noise, less angle dependence, greater resolution of small vessels, and no aliasing.^31,32 Furthermore, power Doppler can be reconstructed into a 3D volume, which, as compared to magnetic resonance angiography (MRA), can simultaneously demonstrate arteries and veins, include small-caliber vessels.^33–35 Power Doppler, however, sacrifices information about flow direction and velocity.³¹ The power Doppler signal also tends to be visible outside the boundaries of blood vessels and, as such, vessels appear larger on power Doppler imaging than they do on MRA.³³ Furthermore, the high sensitivity of power Doppler may result in visualization of small vessels of limited relevance, thus diminishing intraoperative utility.³³ Finally, power Doppler is limited by operator dependence, motion sensitivity artifacts, and overall poor resolution compared to other imaging modalities.^33,36

Contrast-enhanced Ultrasound

Contrast-enhanced ultrasound (ceUS), which has been used extensively in liver^37–39 but is relatively new to neurosurgery, is another modality that allows real-time visualization of tissue vascularity. Therefore, ceUS is useful for the identification of tumors that recruit an avid vascular supply.⁴⁰ Unlike CT or MRI contrast agents, ceUS contrast agents, which are composed of small gaseous microbubbles that resonate when struck by US waves, are transported through small capillaries without diffusion into the interstitium and lack major side effects or toxicities.^41,42 ceUS augments the simplicity and workflow integration of 2D US with additional data, in the form of arterial and venous phase duration, peak signal, and magnitude of contrast enhancement, that can help the surgeon better identify tumor, predict vascularity, assess grade, and navigate around vascular structures.^40,42–44

The most recent generation of contrast agents and the development of contrast-specific sonography algorithms have enabled US imaging with improved spatial and temporal resolution.⁴⁵ Furthermore, image quality is unaffected by angle of insonation.⁴⁴ Unlike Doppler imaging, ceUS can simultaneously show high- and low-flow vessels and provide more detailed information regarding tumor microcirculation and perfusion dynamics.^40,43,44 Unfortunately, the expertise required to optimize imaging parameters and the requirement that a single view be maintained throughout contrast administration can introduce significant inter-operator variability, decrease the flexibility of tumor visualization, and require the use of specific high-end US equipment.⁴⁴ Furthermore, the intraoperative nature of the contrast agent may require that intraoperative imaging occur prior to coagulation of tumor feeding vessels, thus altering the operative workflow.⁴⁴ Finally, there are no US Food and Drug Administration-approved contrast agents approved for neurosurgical use or standardized approaches for tumor resection in the United States at this time.⁴²

Ultrasound Elastography

Elastography is a modality-independent imaging modality that maps elastic properties of tissue. In general, the strain, or relative deformation, of a portion of tissue can be related to stress, an applied or observed force per area, as a function of the tissue’s stiffness. Ultrasound-based elastography was first described in 1991 and has since been applied in a variety of surgical and diagnostic contexts including diagnosis of prostate cancer, assessment of liver fibrosis and evaluation of breast masses.^46–49 Intraoperative 2D elastography of brain tumors has been generated by two methods: active compression with low frequency axial probe vibration (vibrography) and passive observation of arterial pulsations.^50,51 3D US elastography has also been reported.⁵² Although US elastography has negligible acquisition and computation times, its use is hindered by uncertain correlation with histopathology, high levels of noise, and significant operator dependence in image acquisition and interpretation.⁵³

Endonasal Ultrasound

A variety of US transducers have been applied to endoscopic transsphenoidal surgery. Early approaches, developed and characterized in the 1990s, utilized novel, relatively high frequency (10–15 MHz), rigid US transducers that could be advanced into the sphenoid sinus and acoustically coupled to the pituitary with saline.^54,55 The size of these early transducers, however, limited their ability to provide adequate target visualization and to avoid artifacts from the sphenoid bone.⁵⁶ More recently, however, rigid transsellar probes developed specifically for the needs and constraints of pituitary surgery have been described.⁵⁷ Furthermore, novel endoscopic and rigid probes with side-facing transducers have enabled intrasellar visualization.^56,58 Although these side-facing probes offer superior resolution and fewer artifacts, their utilization is limited by the need to create a resection cavity prior to probe insertion and by unfamiliarity with nonintuitive imaging planes.⁵⁷ 3D US reconstructions, which may improve the surgeon’s ability to incorporate US imaging, has not been described for either type of transducer.

Ultrasound Applications

One primary goal of brain tumor surgery is to remove as much of the tumor as possible while minimizing post-operative neurological morbidity. Intraoperative navigation has become standard of practice in many institutions for initial localization and assessment of tumor margins during resection. The utility of preoperative imaging for intraoperative navigation to achieve these ends, however, can be limited by the well-characterized phenomenon of brain shift, in which structures change shape and position as a result of intraparenchymal swelling, gravity, tumor resection, CSF drainage, and other factors. As such, there is a need to incorporate intraprocedural measurements of brain shift into the neuronavigation process in order to update the interpretation of preoperative imaging to reflect intraoperative conditions. Various approaches, including intraoperative MRI (iMRI),^59,60 cone-beam CT,⁶¹ biomechanical modeling and computation,⁶² and stereo cameras⁶³ have been proposed as solutions to this problem. Compared to other imaging modalities, the use of intraoperative US (ioUS) is attractive due to its low cost, minimal interruption of the operative flow, and lack of radiation exposure. In this context, US has three major applications: (1) intraoperative navigation, (2) assessment of extent of resection, and (3) brain shift monitoring and compensation.

Intraoperative Navigation

The utility of an intraoperative imaging modality can be assessed by its accuracy and precision in localizing tumors and reliability in differentiating tumor from surrounding structures. Each of the varieties of US has notable strengths and weaknesses for these applications.

2D US shows significant correlation with MRI in assessing preoperative tumor volume, and gauging extent of resection in primary and recurrent gliomas as well as metastatic tumors.⁶⁴ Sonographic features can help differentiate low-grade gliomas, which can demonstrate calcific features and mild hyperechogenicity, from high-grade gliomas, which can show changes due to necrosis.⁶⁵ Intraoperative 2D US provides greater fidelity in primary resection cases compared to recurrent tumors with prior surgery and/or radiation,^64,66 possibly due to the increased echogenicity of edema and post-radiation gliosis and necrosis.⁶⁷ When used in conjunction with MRI, US improves identification of tumor bulk beyond the margins visualized on gadolinium-enhanced and non-gadolinium-enhanced T1-weighted MRI and helps differentiate tumor from edema visible on T2-weighted imaging.⁶⁸ Furthermore, US is comparable to CT in guiding stereotactic brain biopsy, as assessed by rate of diagnostic yield.⁶⁹

The utility of 3D US was first assessed in a series of 28 operations for primary and metastatic brain tumor resection.⁷⁰ Comparison of perceived tumor on 3D US with pathologic diagnosis of biopsies taken from the observed tumor border revealed a 74% concordance for low-grade astrocytomas, 83% for anaplastic astrocytomas, 77% for glioblastomas, and 100% for metastases. Notably, the ability of intraoperative US to accurately delineate the histopathologic tumor border was equivalent to that of T2-weighted MRI and better than that of T1-weighted MRI.

Intraoperative Doppler imaging has been used to identify the location of blood vessels in real time during brain tumor resection. In a series of glioma resections, tracked 2D Color Doppler imaging was used to landmark major vascular structures onto preoperative imaging, although the presence of brain shift limited the utility of this integration.⁷¹ Tracked 2D Power Doppler imaging has been used to localize and guide the surgical approach to resection of hemangioblastoma, many of which are not visible on standard 2D US.³⁶ Finally, reconstructed 3D Power Doppler images have enabled intraoperative navigation around major vascular structures without reliance on preoperative imaging that does not account for brain shift.^33,72

Contrast-enhanced US provides reliable intraoperative visualization of peritumoral and intratumoral vascularity.⁷³ Strong contrast enhancement is seen in meningiomas, hemangioblastomas, and metastases, whereas less avid enhancement patterns are observed in gliomas and lymphomas.⁷³ Furthermore, the magnitude of ceUS contrast enhancement correlated with the extent of vascularity observed on digital subtraction (DS) angiography but not with the degree of contrast enhancement observed on CT or MRI. This finding likely reflects the fact that CT and MRI contrast agents, which diffuse into the interstitium, highlight areas of disrupted blood-brain barrier in addition to regions of increased vascularity. In a study of seven patients with primary or metastatic central nervous system (CNS) neoplasms, a single contrast bolus produced a hypoechoic appearance for peritumoral edema compared to a hyperechoic appearance in tumor.⁷⁴ In a larger series of 71 brain tumor patients, glioblastomas were characterized as largely heterogeneous with marked enhancement and rapid arterial and venous phases while low-grade gliomas were contrast enhancing with slower vascular phases.⁴⁰ Finally, attempts to quantify observed perfusion parameters, and therefore standardize diagnostic information gleaned from ceUS, have been described.⁴²

US elastography is still in early stages of development and characterization with regards to brain tumor resection. Arterial pulsation-based US elastographic approaches have identified strain as a mechanical property that can be used to differentiate tumor from normal parenchyma, although it should be noted that these images have only been compared with 2D US and not MRI or histopathology.^53,75 3D Elastography approaches have supported these general observations and further suggested that axial and shear strain data can be used intraoperatively to predict the location of dissection planes during resection.⁵² Finally, in a study of a variety of brain tumors, US vibrography demonstrated that tumor had distinct stiffness properties as compared to normal parenchyma but failed to identify correlation of these patterns with tumor grade or type.⁵⁰

Endonasal US is an emerging technology with unclear significance in the context of transsphenoidal tumor surgeries. Studies conducted with the early rigid endonasal US probes suggested that US could identify pituitary tumors with 81% sensitivity.⁵⁴ Furthermore, transsellar US has been used to identify adenomas that could not be imaged on MRI.^55,57 In a recent case series of 9 patients, however, a narrow probe with a side-facing transducer and power Doppler capability was able to provide intrasellar visualization of pituitary macroadenomas with a higher resolution (0.19mm x 0.22mm) than that available via T1-weighted MRI (1mm x 1mm) and without the need to remove extra bone to accommodate the probe.⁵⁶ A subsequent case report that used a similarly small endoscopic transducer revealed consistent visualization of tumors and surrounding anatomic landmarks and vasculature.⁵⁸ Transcranial US through a frontal burr hole has also been used as an adjunct in transsphenoidal resections of pituitary macroadenomas and, in contrast to endonasal probes that compete with space required for surgical instrument access, provided visualization during the resection itself.⁷⁶

Assessment of Extent of Resection

Extent of resection (EOR) contributes to improved patient survival in a number of CNS tumors, including high-grade glioma, low-grade glioma, and high-grade meningioma.^77–81 However, surgeon assessment of extent of resection is liable to error, even by experienced clinicians.⁸² As such, reliable methods of assessing extent of resection intraoperatively can serve as critical adjuncts. Ultrasound offers a promising venue given its flexibility and real-time feedback.

While 2D US can provide real-time localization of tumor prior to resection, its ability to reliably identify residual tumor is limited by a variety of factors. In one study, samples taken from tumor margins, as delineated by US, were found to be in accordance with histopathology in 84% of sites that appeared tumor positive and 88% of sites that appeared tumor negative.⁸³ The high number of false positives was thought to be a result of acoustic enhancement artifact from saline and clotted blood in the resection cavity, both of which can appear hyperechoic on US.⁸³ A previous study of intraoperative US in glioma resection reported 89% concordance with histopathology in hyperechoic areas that clearly extended into iso-echogenic brain parenchyma but only 56% concordance along the hyper-echoic rim of the resection cavity.⁸⁴ Comparison of ioUS and 1.5T intraoperative MRI in 26 patients with various brain tumors found that 2D US reliably detected residual tumor greater than 1 cm, but lost fidelity for smaller remnants.⁴ Notably, in 2 cases where residual tumor was suspected on ioUS but not MRI, pathologic confirmation revealed absence of tumor. These results supported an earlier comparison of intraoperative 2D US with 0.2T intraoperative MRI, which also identified greater difficulty detecting residual tumor on ioUS compared to MRI.⁸⁵

Investigation of intraoperative 3D US has revealed limitations similar to those of 2D US. Examination of 3D US for resection of glioblastoma in 19 patients revealed a diminishing accuracy of ultrasound tumor detection with each stage of surgery: the sensitivity and specificity prior to resection of 95% and 95%, respectively, declined to 88% and 42% during resection, and 26% and 88% following resection.⁸⁶ Significant false positive detection of tumor during resection may be attributable to the confounding effects of edema and boundary artifacts. A study of 3D US- and 5-aminolevulinic acid (5-ALA)-based glioma resection concluded that 3D US and 5-ALA together enhance extent of resection and 3D US is particularly important in the resection of non-enhancing gliomas.⁸⁷ Rates of gross total resection of primary brain tumors as guided by intraoperative 3D US vary significantly, with applicability yet to be determined.^66,88–90

High-frequency US offers superior intraoperative tumor detection ability compared to that of 2D US. In primary and recurrent surgery for glioblastoma (GBM), hfUS was reported to have a higher sensitivity for tumor detection (76%), as determined by histopathology from navigated biopsies, than either or 2D US (24%) or gadolinium-enhancing T1 iMRI (55%).²⁹ The specificity of hfUS (58%), however, was lower than that of 2D US (96%) or gadolinium-enhancing T1 iMRI (74%). In a similar analysis of surgery for low-grade gliomas, hfUS was reported to have a sensitivity (79%) greater than that of 2D US (21%) and comparable to that of T2/FLAIR iMRI (83%).² The specificity of hfUS (67%), as with GBM, was lower than that of 2D US (100%) and comparable to that of iMRI (67%). In both studies, hfUS demonstrated fewer surgically-induced artifacts and, therefore, fewer inconclusive assessments of residual tumor presence. hfUS has also been applied intraoperatively to assess extent of CNS tumor resection, with one study demonstrating a 95.5% gross total resection rate as verified by post-operative MRI.³ Instances of missed residual tumor may have been a result of diminished resolution of hfUS at superficial (<1 cm) depths or incomplete scanning.

ceUS offers additional data on vascularity to help differentiate tumor from edema during intraoperative assessment of extent of resection. In a series of 10 patients with GBM, ceUS was able to identify residual tumor in 9 cases, as confirmed by histopathologic analysis.⁴⁴ One case with diminished contrast-enhancement was attributed to partial devascularization during the course of the resection. In a series of 120 patients with gliomas, intraoperative ceUS demonstrated 62% sensitivity and 93% specificity in detecting residual tumor, as evaluated by pathologic analysis of biopsy specimens around the resection cavity.⁶⁷ False positive detection of residual tumor was particularly notable in cases of recurrent glioma, with prior radiotherapy, suggesting that gliosis poses a significant barrier to evaluation with ceUS.

Endonasal US offers the potential for intraoperative EOR assessment during pituitary surgery, although data is still largely lacking. In a study of 24 pituitary adenomas, use of a transsellar endonasal US probe with front-facing transducer was able to identify residual tumor and improve EOR in 4 operations.⁵⁷ Residual tumor detection with side-facing transducers have also been reported.⁵⁶ Although it is difficult to generalize these early results, the success of iMRI for the determination of EOR in pituitary adenoma surgery suggests that further evaluation of endonasal US imaging is warranted.^91,92

Brain Shift Monitoring and Compensation

In addition to providing real time visualization, intraoperative US can also be used to measure brain shift over the course of surgical resection. Observed shifts, which can be represented in the form of a deformation field, can be used to update preoperative imaging (usually MRI), which generally offers superior resolution and tissue differentiation, to reflect the current intraoperative state. The generation of a deformation field relies on registration, the algorithmic alignment of two images of the same anatomy, such as those obtained before and during a resection. Registration strategies vary, but it is worth drawing a distinction between rigid registration, which maps two images with only translation and rotation, versus non-rigid registration, in which the images can also be scaled or otherwise deformed. In general, non-rigid registrations may more accurately represent observed shifts at the cost of increased computation time and more variable and uncertain algorithm behavior as compared to rigid registrations.⁹³ Work conducted at our own institution has demonstrated that the uncertainty associated with a given registration can be modeled and visualized in a way that is useful to the surgeon.⁹³

Because of its low cost, safety, and speed, US is an attractive candidate for this application. Registration of US to MRI, however, is technically challenging, largely due to the fact that the two modalities highlight different types of anatomic details and contain different artifacts and noise patterns.⁹⁴ Accordingly, many registration algorithms that rely on overall image similarity, such as mutual information, sum of squares difference, and correlation ratio, fail to produce satisfactory results when registering US to MRI.⁹⁵ Several other approaches to merging intraoperative ultrasound with anatomic registration have been proposed but await validation in larger cohorts of patients. These include a probabilistic function that matches hyperechoic structures for rigid registration,⁹⁶ a Bayesian registration framework that incorporates local region information to improve the compensation for missing tissue and non-stationary image characteristics,⁹⁷ generation of a “pseudo-US” image from MR,⁹⁸ and registration of Doppler US to preoperative MR angiography.^94,99

Approaches to the registration of preoperative MRI with pre-durotomy US vary in complexity and in accuracy. One approach used neuronavigation to register preoperative MRI with pre-durotomy US and demonstrated accurate co-registration of reference structures, such as the choroid plexus and the falx, but not of cortical and peri-tumoral points, which are affected by brain shift.¹⁰⁰ Rigid registration approaches have demonstrated more accurate alignment of cortical features than neuronavigation alone. In a study of brain tumor resection, registration of pre-durotomy 3D US to preoperative MRI determined brain shift to be 3.0 mm parallel to the direction of gravity and 3.9 mm perpendicular to the direction of gravity.¹⁰¹ Another study, in which pre-durotomy US was rigidly registered to preoperative MRI with a novel implementation, demonstrated a landmark-based error of 2.52 mm in a series of 14 patients.¹⁰² These observed misalignments may result from either true brain shift or calibration and registration errors.¹⁰³ The magnitude of these errors, which can be predominately attributed to fiducial mismatch when aligning MRI and to imperfect US probe calibration, is estimated to be about 1–1.5 mm.¹⁰¹

Pre-durotomy 3D US has, in one study, been used to register preoperative MRI to patient coordinates in the operating room instead of standard skull-based fiducial registration. 3D US-based registration yielded more accurate registration of intra-cranial landmarks, such as the anterior and posterior commissure, at the cost of diminished accuracy in the registration of external anatomical landmarks.¹⁰⁴ As the purpose of the fiducial-less registration was to match internal image features, these results suggest that the method is successful and offers advantages over registration based on external anatomy or could be a useful adjunct.

Post-durotomy 3D US has also been registered to preoperative MRI. Using a rigid mutual information algorithm, as noted above the authors documented brain shifts of 3.0 mm parallel to gravity and 3.9 mm perpendicular to gravity before durotomy and 3.2 mm and 5.3 mm, respectively, after durotomy.¹⁰¹ Interestingly, in this study, the component of shift that occurred post-durotomy was smaller in magnitude than pre-durotomy shifts, which, as previously discussed, may represent either calibration errors or true shift.

Registration of pre-dural opening to post-resection 3D US volumes, perhaps the most useful assessment of brain shift in the context of maximizing EOR, is unfortunately complicated by the resection of tumor and intraoperative artifacts. A single study implemented and evaluated this method in 16 cases of brain tumors and reported that both rigid and non-rigid registration decreased distances between labeled anatomic points from their original positions.¹⁰⁵ The mean shift reported in the study at subcortical sites was 3.2 mm.

Future Directions

Neurosurgical applications of ultrasound have benefited from the development of new ultrasound transducers, contrast agents, and processing systems, largely driven by other medical specialties with higher clinical volumes. Furthermore, improved beam forming and evolving image processing systems, inspired by mass commercial applications, will likely also translate to improvements in US technology. Refinement in intraoperative US image quality and increasing native 3D acquisitions may overcome challenges with tissue differentiation and allow for more accurate assessment of extent of resection. Additionally, closer integration of US with neuronavigation image processing will offer surgeons the ability to assess EOR with multiple, brain shift-corrected imaging modalities. Further research will be necessary to characterize and correct artifacts that currently limit the use of US in neurosurgical applications and to clarify the value of newer US modalities for achieving maximal safe extent of resection.

Conclusions

Brain tumor resection continues to strive to minimize disease burden while preserving neurologic function. Ultrasound is able to provide real-time intraoperative feedback to guide surgical resection. Mounting data points to the reliability of ultrasound detection of tumor with pathology. However, the challenges of using ultrasound in a neurosurgical setting, including artifacts that accrue in the resection cavity and intra-operative variability in image production, continue to limit widespread adoption of ultrasound technology. Improved sensitivity and specificity of identifying residual tumor during resection, integration of real-time ultrasound feedback with anatomic imaging, and compensation for intraoperative brain shifts may be offered by a number of novel ultrasound modalities, including 3D reconstruction, in the future. Studies to evaluate the impact of these developing technologies on clinical outcomes following brain tumor resection in large patient series will continue to inform the utility of ultrasound technologies.