Preoperative functional MRI (fMRI) is limited by a muted BOLD response caused by abnormal vasoreactivity and resultant neurovascular uncoupling adjacent to malignant brain tumors. We propose to overcome this limitation and more accurately identify eloquent areas adjacent to brain tumors by independently assessing vasoreactivity using breath-holding and incorporating these data into the BOLD analysis.
Local vasoreactivity using a breath-holding paradigm with the same timing as the functional motor and language tasks was determined in 16 patients (9 glioblastomas, 1 anaplastic astrocytoma, 5 low grade astrocytomas, and 1 metastasis). We derived a model based on coherence for analyzing BOLD fMRI that takes into account the altered hemodynamics adjacent to brain tumors.
Activation maps computed using the coherence model were overall similar to standard activation maps. However, the coherence maps demonstrated clinically meaningful areas of activation that were not seen using the standard method in 12/16 cases. This included localization of language areas adjacent to brain tumors, where the coherence method results were confirmed by intra-operative direct cortical stimulation. Enhanced task response maps based on vasoreactivity mapping demonstrated more robust, anatomicallycorrect activation, in particular adjacent to tumors as compared to maps obtained without vasoreactivity information.
The present preliminary results demonstrate the principle that the neurovascular uncoupling known to affect the accuracy of BOLD fMRI adjacent to brain tumors may be, at least partially, overcome by incorporating an independent measurement of vasoreactivity into the BOLD analysis.
Neurosurgical resection remains the most important treatment option for malignant brain tumors as both the length and quality of survival are improved with maximized tumor resection [1,2]. Therefore, the goal of brain tumor surgery is to maximize the resection of the tumor while avoiding important adjacent eloquent cortices, whose inadvertent resection can lead to devastating neurological consequences. In order to preserve vital neurological function located near the tumor, it is important for the neurosurgeon to be able to identify the anatomical location of the eloquent cortices, either pre- or intraoperatively.
Traditionally, the eloquent cortices, such as the motor cortex, have been identified by direct cortical stimulation. More recently, blood-oxygen-level-dependent functional MRI (BOLD fMRI) has been used successfully in planning and carrying out the resection of brain tumors [3,4]. However, it has been found that BOLD maps adjacent to brain tumors have limited accuracy [5- 9]. BOLD fMRI is based on the premise that there is a coupling between neuronal activity and blood flow. However, the neovasculature of malignant tumors is known to be abnormal both structurally and functionally , including changes in vascular reactivity [10-14]. Since BOLD fMRI measures the vascular response (rather than neuronal activation), in patients with malignant brain tumors the abnormal tumor neovascularity does not respond as vigorously to increased neuronal activity, which leads to a muted BOLD response . A number of recent studies have correlated the anatomical location of neurovascular uncoupling, defined by quantitative measurements of cerebrovascular reactivity, with false negative BOLD fMRI results [7,16-18].
As a consequence, models assuming a uniform rather than an abnormal hemodynamic response function over the whole brain may not be sufficient to detect activation in an eloquent cortex influenced by the tumor. Data analysis based on the uniform hemodynamic response functions may lead to missed activation (or false negative errors) in the BOLD statistical parametric maps.
A possible way to overcome the limitation of false negative results caused by the presence of abnormal tumor neovasculature and resultant neurovascular decoupling is to independently measure vasoreactivity and to incorporate these data into the BOLD fMRI analysis. One way to independently measure vasoreactivity is by breath holding, which leads to hypoxia and hypercapnia, which under normal circumstances leads to vasodilatation of the cerebral vasculature [14,17,18]. These changes in vasodilatation can be measured by fMRI in the same way as BOLD fMRI responses to routine task paradigms are measured. The assumption is that the hemodynamic response function of abnormal neovasculature is altered from the norm in the same way irrespective of the stimulus (breath-holding or task fMRI). If this assumption holds true, then one should be able to detect BOLD activation even in areas of abnormal neovasculature by searching for coherence between the hemodynamic response function in the breath-holding and routine task fMRI paradigms.
The purpose of this study is to derive a voxel-specific model that takes into account altered hemodynamics by comparing BOLD response maps obtained during motor and language tasks with and without the incorporation of a breath-holding task. We hypothesize that BOLD response maps to motor and language tasks computed this way will be overall similar to standard activation maps based on uniform hemodynamic response models but will also show new areas of activation that accurately reflect increase in neuronal activity that were not detected by the standard method of BOLD fMRI analysis due to altered hemodynamics of the tumors.
MATERIALS AND METHODS
The research performed was in full compliance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). The institutional review boards of Memorial SloanKettering Cancer Center and Weill Cornell Medicine approved the study. Informed Consent was not obtained, as the study was a retrospective review.
Each patient underwent an anatomical brain MRI and an fMRI as part of their routine preoperative care. All underwent resection of the tumor within two days of the MRI. Pathological examination of the resected tumors revealed the diagnosis of glioblastoma multiforme (N = 9), anaplastic astrocytoma (N = 1), low-grade glioma (N = 5), or metastasis (N = 1). See Table 1 for demographic data.
fMRI and anatomical MRI scanning
In total, 18 fMRI scans were performed on 16 patients. Patients also underwent an anatomical brain MRI on a 3.0 Tesla GEMS (Waukesha, WI) clinical MRI system with an 8-channel head coil. Breath-hold and task-specific fMRI were performed using echo-planar fMRI (TR = 4 s; TE = 40 ms; 90o flip angle; 128 × 128 matrix; 240 mm FOV; 4.5 mm slice thickness).