2. Functional magnetic resonance imaging (fMRI) is increasingly
used in the work-up of patients at the preoperative stage to assess the
relationship between the functionally eloquent cortex and brain pathology. Inter-individual
normal variations of anatomy render such assessment unreliable based
on structural imaging alone despite the definition of clear anatomical landmarks.
This is even more of an issue when normal anatomy is obscured by a tumour mass
effect or when functional anatomy is altered due to cortical plasticity.
fMRI has seen a rapid evolution from its first human application in 1991 to an
essential tool in the exploration of human brain function, most prominently in the
scientific arena.
In 2007, new Current Procedure Terminology (CPT) codes were developed for fMRI
by the American Medical Association, signifying the transition of fMRI to a
valuable tool in a clinical setting.
Recent major advances in clinical fMRI make its acquisition, image processing, and
even integration of its findings for neuronavigational purposes relatively easy.
However, the technique is not without limitations and validation issues which are
easily forgotten when color activation maps become readily available at the single
click of a button. The theoretical background, the validity in brain tumour
patients, and several considerations of fMRI are addressed.
3. fMRI Background:
Blood Oxygenation Level-Dependent (BOLD) fMRI is the most commonly used
functional MR neuroimaging technique. BOLD fMRI takes advantage of the tight
link between local neuronal activity and blood flow, called neurovascular coupling
due to neurovascular coupling, blood flow and volume increase locally with an
increase of neuronal activity. This leads to an increase in oxygenated blood that is
disproportionate to the increased need of oxygen for neuronal activity. As a result,
there is a relative decrease of paramagnetic deoxygenated hemoglobin which in
turn leads to an increase of MR signal in those areas of the brain that are active.
Such signal changes are small and relative, which means that many
measurements need to be made, typically during an alternation of active and
baseline conditions in a task that aims to activate the functional brain region of
interest. Furthermore, the signal changes occur at a delay after and are more
prolonged than the neuronal activity, defined by the hemodynamic response
function. A statistical model is created to assess the correlation of the measured
signal changes with the task, taking the hemodynamic response function into
account. The resulting statistical map is threshold at a certain p or T value and
overlaid in color on a high-resolution anatomical image which is acquired
separately. This is the typical color “activation” map produced by an fMRI image
processing software, which is merely a combination of anatomical and statistical
information very indirectly representing neuronal activation.
4. Task-Based fMRI
For clinical application, almost exclusively task-based fMRI is used. During
the performance of a task by the subject in the scanner, rapid imaging of
the brain is performed. Typically, the entire brain is scanned at intervals of
3–5 s for a duration of about 5 min so that > 100 measurements are made
per task. The task consists of active and baseline conditions, which
commonly alternate in blocks of 20–40 s. Such so-called blocked
paradigms are statistically robust, since a lot of signal is acquired for each
condition, but they are restrained because they do not leave much room
for unexpected or short stimuli. In an event-related task design, individual
stimuli, each representing a specific condition, are presented in random
order and rapid succession. Such a task design offers the possibility to
present unexpected stimuli as well as many different condition. Rendering
it very flexible, but statistically less robust since the signal acquired per
condition is generally low. For clinical application, blocked designs are
generally well-suited and preferable.
5. Clinical Pre-Surgical fMRI Studies:
The aim of neurosurgery in brain tumour patients is maximum tumour resection, while
at the same time minimizing the risk of new functional deficits post-operatively. For
optimal results, the relationship between the tumour margins and eloquent brain
regions needs to be established as accurately as possible. The gold standard for such
assessment is intraoperative ECM, which has in fact been shown to significantly modify
long-term survival in low-grade glioma patients. However, intraoperative ECM is
invasive, requires experience and expertise of the neurosurgical team, increases
surgery duration, and requires awake and active participation, collaboration, and
motivation of the patient. Additionally, only a limited number of tasks can be tested.
Functional MRI may be used to make a risk assessment preoperatively, which is of
particular value in young low-grade glioma patients, to plan and guide the
neurosurgical approach, shorten surgery duration, and obtain prognostic information
prior to surgery. For fMRI to be used in such a setting, both high sensitivity and high
specificity are required. High sensitivity for eloquent brain regions is needed to reduce
the false negative rate so that no eloquent cortex is missed and no functional deficit is
induced by surgery. At high specificity the false positive rate is low, which means that
the visualized areas of activation relate to truly eloquent or critical brain regions. At
low specificity non-critical brain regions are also visualized, inducing the risk that such
areas are avoided at surgery and are subsequently exposed to a less extensive
resection than would have been possible.
6. Motor Function:
Motor cortex assessment has been validated in a multitude of studies
that generally report a good correlation between fMRI and
intraoperative ECM. Reported sensitivities for localization of the primary
motor cortex range from 88–100 % . Specificities are also high, ranging
from 87–100 % . Such high reliability may be contributed to by the robust
activation that is seen with simple motor tasks that can be easily
performed by the majority of patients. Also the functional anatomical
stability of the sensorimotor area, at both the macroscopic and
microscopic levels, probably contributes to the reliability of fMRI of
motor function . Spatial accuracy of fMRI motor cortex localization is
found to be within the range of 1–2 cm. Yetkin et al reported that all
intraoperative ECM and fMRI sites of activation were within 2 cm, while
87 % were within 1 cm. Importantly though, reliability seems to be
decreased with high tumour grades, as demonstrated by Bizzi et al. In
benign and malignant brain mass lesions in or near the primary motor
cortex, overall sensitivity was 88 %, but only 65 % in grade-IV gliomas.
7. Language Function:
In contrast to the high validity shown for fMRI of motor function, results
from language function validation studies are controversial, varying from
100 % sensitivity for fMRI to identify all critical language areas to as low
as 22 %. Reported specificity is even more variable, ranging from 0– 100 %.
Validation studies of fMRI of language function in brain tumour patients
are relatively scarce, are generally performed in small patient populations,
and suffer from differences in the validation methods used among the
studies, disparities of brain lesions, and the variety of the language tasks
performed preoperatively and during intraoperative ECM. The mapping of
language function is also more complex than that of the sensorimotor
cortex due to the lack of consistent surface landmarks and substantial
inter-individual variability. The language cortical network seems to consist
of critical regions, essential for language processing, and participating but
non-critical areas, which may be resected without inducing a permanent
language deficit. These areas cannot be reliably distinguished with fMRI,
resulting in low specificity of fMRI compared with intraoperative ECM.
8. Other Functions:
The visual cortex has been a frequent topic of study since the early days of
fMRI, due to its relatively strong BOLD response and easy implementation
of stimulus paradigms. Presurgical mapping of the primary visual cortex
has been described and may be indicated when the normal anatomy is
severely distorted by the tumour and/or when the brain structure of
interest is located deep inside the brain and cannot be assessed by ECM.
Commonly used stimulus paradigms are flashing lights presented with
light-proof goggles and reversing black-and-white checkerboards .
Another function that may be assessed with fMRI is visuospatial attention,
failure of which results in spatial neglect.
This condition arises with damage of the temporoparietal or frontal
cortex, the thalamus or the basal ganglia, generally of the right
hemisphere. It is an invalidating condition, in which patients behave as
if the left part of the world does not exist. Functional localization may be
assessed with fMRI using a line bisection task, in which patients are asked
to bisect 20-cm horizontal lines, which has been used successfully during
ECM.
9. Critical Issues:
There are several limitations of fMRI that need to be considered when using
the technique for pre-surgical assessment of brain tumour patients. These
include issues that are inherent to the technique, such as spatial and
geometric uncertainty, tumour effects on the BOLD signal, inter- and intra-individual
variability, lack of discrimination between essential and modulating
brain regions, and lack of information on the underlying white matter. The
imaging sequence used for BOLD fMRI is particularly sensitive to
postoperative effects, such as metallic implants and surgical staples, air
underneath the skull flap, and blood products, as well. This means that
additional care needs to be taken in patients who have had previous surgery,
biopsy, or hemorrhage. Small regions of hemosiderin deposition may not be
visible on conventional imaging, but will show large artifacts in the BOLD
fMRI data. Such artifacts are obscured on the fMRI activation color maps,
which would simply show decreased or no activation in the artefactual area.
It is therefore crucial that the raw data are scrutinized for such artifacts in
every patient. Other issues with pre-surgical fMRI, that may be resolved in
the future, are the lack of standardization of tasks and image processing
techniques.
10. Four resting state networks in a normal subject. Also shown are corresponding time courses of
the independent component coefficients.(Colors quantify the z-value of correlation of the time
series with the corresponding ICA coefficient. Either positive or negative z-values are shown.)
11. Activation patterns in response to visual stimuli. Note the lack of
activation in the central visual cortex when using projection screen.
12. Visual area locations. A, Parasagittal anatomical image from one subject
indicating the slice selection (perpendicular to the calcarine sulcus) for the
retinotopy measurements and the control conditions. B–D, Visual areas V1, V2,
V3, V3A, and V4v depicted on in-plane anatomies from a similar location (near the
middle of the 8 in-planes) in three control subjects. E–G, Visual areas depicted on
in-plane anatomies from a similar location in three dyslexic subjects.
13. Location of area MT1. A, Parasagittal anatomical image from one subject indicating the
slice selection (parallel to the calcarine sulcus) for the moving dots and the test conditions.
The blue lines were slices that contained the MT1 region of interest (ROI) in this subject. B–
D, Brain activity in one slice containing the MT1 ROI from three control subjects. Reddish
voxels show regions with greater response to moving versus stationary dot patterns.
Images were chosen to optimally show MT1 in the right hemisphere (arrows), although
activity from left hemisphere MT1 and V1 are also present in some cases. The image in B is
from the same brain as the sagittal image in A and corresponds to the most inferior of the
three blue slices. The MT1 ROI was defined by outlining (dotted line in B) the strongest
area of activity that was approximately lateral to the junction between the calcarine sulcus
and the parieto-occipital sulcus, and beyond the retinotopically organized visual areas.
E–G, Slices with MT1 ROIs in three dyslexic subjects.
14. This image is in
radiological format
and shows an
auditory language
map averaged over
40 subjects.
There is bilateral
primary auditory
cortex (yellow
arrows), left
Wernicke’s (green
arrows) and Broca’s
(red arrows), and
SMA (orange arrows).
In this auditory task,
the subject hears
phrases, such as “The
funny guys at the
circus” and thinks
in his/her mind of a
word that fits.
16. Figure 4. Functional reorganization of motor area was observed in the patients
with short term and long term spinal cord injury (sci).
Figure 4a. Control.
Figure 4b. Two years after sci.
Figure 4c. More than 10 years after sci.
17.
18.
19.
20. Localization of primary motor cortices in 28 years old patients with brain tumor:
Note that the posterior and medial displacement of motor cortices by the mass lesion
21. Localization of language cortices in 28 years old patients with brain tumor:
Note that the strong left lateralization of language activities.
25. fMRI in preoperative planning.
3a. Preoperative planning for brain biopsy in a patient with a lesion (arrow) in
the left frontal lobe (Broca’s area). Significant bilateral language activation
was observed in this patient.
3b. fMRI was used to determine the language area and its relation to the tumor
for a patient with glioblastoma (arrow).
26.
27.
28.
29.
30.
31.
32. Patient K, male, 37 years old, with glioblastoma in the left temporoparietal
area,activation from visual task, R - right hemisphere; L - left hemisphere.
33. Patient P, male, 28 years old, with glioblastoma in the left temporal lobe, activation
from visual and auditory tasks, R - right hemisphere; L - left hemisphere, WA –
activation in Wernicke's area. Red - activation elicited by visual task, blue - activation
elicited by auditory task, purple - activation that coincided in two tasks.
34. Patient KO, female, 53 years old, with glioblastoma in the left parietal area,
activation from visual task (red), R - right hemisphere; L - left hemisphere.
35. Preoperative (A) activity showing the focus
of activity in 4 patients obtained from the
comparison of the visuomotor conditional
blocks of trials with the motor control
blocks of trials and postoperative (B)
anatomical images. A: Functional MR
images show the focus of activity within
the rostral part of the premotor cortex in
the lateral view (in all patients) and in the
top view (in Case 2) of the cortical surface
rendering in standard stereotactic space of
the left hemisphere of each patient. The
gray shaded area of the cortical surface
shows the tumor location. Each blue cross
indicates the location of the preoperative
activity increase in the PMdr. The color
scale indicates the t value range. B:
Magnetic resonance images show the
cortical surfaces as renderings (in standard
stereotactic space) of the left hemisphere of
each patient based on the postoperative
MR image. The deformation of the cortical
surface shows the extent of the tumor
resection. The data demonstrate that, in
each patient, the region involved in the
visuomotor conditional function was not
removed during the tumor resection.
36. Images showing the
preoperative activity
obtained from the
comparison of the
motor control with the
visual control block of
trials. The cortical
surfaces are renderings
(in standard stereotactic
space) of the left
hemisphere of each
patient. The horizontal
yellow line indicates the
dorsoventral level of the
horizontal sections
illustrated in the images
on the right. The
increases in activity
correspond to the
location of the primary
hand motor region in
the precentral
knob of each patient,
37. Tumor heterogeneity Recognition Evaluation on the High Grade Glioma of
the BraTS dataset: Qualitative Example Case BRATS HG0001 a) Gadolinium T1 MRI
Original Image b) FLAIR MRI Original Image c) Overlap between Gadolinium T1 MRI
original image and the Ground Truth masks d) Overlap between Gadolinium T1 MRI
original image and the automatic recognized masks.(Red points are Edema and Cyan
point are the Active Tumor
38. Tumor heterogeneity Recognition Evaluation on the High Grade Glioma of
the BraTS dataset: Qualitative Example Case BRATS HG0001 a) Gadolinium T1 MRI
Original Image b) FLAIR MRI Original Image c) Overlap between Gadolinium T1 MRI
original image and the Ground Truth masks d) Overlap between Gadolinium T1 MRI
original image and the automatic recognized masks.(Red points are Edema and Cyan
point are the Active Tumor)
39.
40.
41. 18-year old boy with a glioma grade 3 (arrows) near the left rolandic area, demonstrating
the importance of fMRI in planning surgery and avoiding postoperative neurological
deficits. fMRI using a visuomotor paradigm showed functional activation less than 2 cm
posterior to the tumour and partial resection was proposed (A,B). Partial resection was
performed using intra-operative cortical stimulation, and followed by application of gene
therapy in the tumour remnant posteriorly. Postoperatively, no motor deficit was present.
42. 26-year old man with right frontal glioma demonstrating the role of fMRI in
deciding on operability. This man was left-handed which increased chances that
language areas are not confined solely to the left hemisphere. fMRI using a verb
generation paradigm showed that the language areas Broca (B) and Wernicke (W)
were strongly lateralized to the left hemisphere because no activation was seen in
the right hemisphere. Complete resection of the right frontal glioma (T) was
performed and the patient did not have any language deficit postoperatively.
43. 59-year old woman with arteriovenous malformation (AVM) in the right parieto-occipital
region, demonstrating risk for postoperative deficit because activation areas are less
than 2 cm of the lesion. fMRI using a visuomotor paradigm showed visual activation
areas (green) medially and caudally relative to the AVM (circle), at less than 1 cm
distance. After endovascular approach with partial embolisation, complete resection was
performed surgically. Postoperatively, a left lower quadrant anopsy was present.
44. (Top row) Acquired axial CT and anatomical MRI slices from the AVM patient.
(Bottom row left) fMRI activation map fused with the anatomical MRI data
showing the Broadmann-17 functional structure and (lower right) corresponding
color-coded (red = left-right; green= anterior-posterior; blue = superior-inferior).
The FA map was used for calculating the tractography data.
45. Recurrent left parietal lobe
anaplastic astrocytoma in 37-
year-old right-handed woman.
Surgery was not initially planned
because of presumed
involvement of receptive speech
area. Left inferior and middle
frontal gyral activation (yellow
arrows) is consistent with
dominant expressive speech area
and is located at anterior border
of more cephalad component of
lesion. Left superior and middle
temporal gyral activation (green
arrows) is consistent with
dominant receptive speech area
and abuts inferior border of
temporal component of lesion,
with superior temporal gyral
activation component lying
anteroinferior to lesion.
46. Preoperative fMRI in patients during contralesional movements. Preoperative anatomic and
functional MRI in Patients 1, 5, and 8 (left column) showed large tumoral infiltration of the
supplementary motor area (SMA) (asterisk), associated with under activity in the SMA in the
pathologic hemisphere, and over activity in the healthy SMA resulting in negative magnitude
laterality indices (LI) (0.40, 0.12, 0.06). Recovery began within the first 3 days and lasted
between 30 and 60 days. In Patients 6, 9, and 11 (right column), tumoral infiltration (asterisk)
was lower, and LI remained positive (0.08, 0.19, 0.12). Arrows indicate predominant SMA
activation. Recovery began after 7 to 21 days, and lasted between 90 and 120 days.
47. Postoperative fMRI. Anatomic and functional MRI in one representative control
and in Patients 4, 6, and 12 during ipsilesional (IL) and contralesional (CL)
movements after surgery. In controls and in patients during IL movements,
supplementary motor area (SMA) (arrow) and premotor cortex (PMC) (arrowhead)
activation predominated in the hemisphere contralateral to the moving hand. In
patients during CL movements, SMA activation was located in the healthy
hemisphere and PMC activation became more important in the ipsilateral healthy
hemisphere. The resection cavity is indicated by the asterisk on axial planes.
48. Axial TIWI in neurologic format show left megalencephaly with an enlarged dysmorphic
left cerebral hemisphere. There is poly microgyria as well as heterotopia. During this
auditory task, there was only activation in the right primary auditory cortex (yellow arrow)
rather the expected bilateral activation. (Color version of figure is available online.)
49. Conclusions.
Functional MRI is a valuable tool in the pre-surgical assessment of
brain tumour patients, but needs to be used with care.
Interpretation of the results requires a lot of experience and may be
difficult. Knowledge of functional brain anatomy is a first
requirement for risk evaluation and to determine which
fMRI tasks need to be performed. The shortcomings of fMRI in a
clinical setting as described above need to explicitly be taken into
account in every patient. In our institution, fMRI, combined with DTI
tractography, is used to aid neurosurgical planning but
intraoperative ECM is always used for confirmation when activation
is shown in the proximity of the brain tumour or if activation is
atypical. Most importantly, the absence of fMRI activation does not
exclude the presence of functional neuronal tissue, not even within
infiltrative tumours.