Evaluation of a Signal Intensity Mask in the Interpretation of Functional MR Imaging Activation Maps

Patients

The 152 studies performed and archived since the introduction of the SIM in June 2000 were reviewed. Medical records on these patients were also reviewed and final diagnoses determined by follow-up or from the results of surgical intervention. Patients were classified, on the basis of their final diagnosis, as having a tumor (primary or metastatic, n = 95), a vascular lesion (arteriovenous malformation, cavernous malformation, infarct, or hemorrhage; n = 42), trauma (n = 2), or other findings (nonlesional epilepsy, visual disturbance, or cognitive decline; n = 13). All patients provided written informed consent allowing us to use their image data for research purposes.

Functional MR Imaging

Imaging was performed using 1.5T commercial scanners (GE Medical Systems, Milwaukee WI) equipped with high-speed gradients. Single-shot EPIs were acquired in the coronal plane at 20–24 section locations. Technical parameters for these images included the following: 64 × 64 matrix, 85° flip angle, TR/TE of 2000/40, 24-cm FOV, 6-mm section thickness, 1-mm gap, and bandwidth of ±62.5 kHz. The number of images and length of imaging varied depending on the paradigm used. Imaging duration ranged from 3 to 5 minutes. Among other diagnostic MR imaging sequences performed in each patient was a high-resolution 3D gradient-echo sequence, a T1-weighted contrast sequences to obtain a whole-brain anatomic dataset (TR/TE of 21/6, 20° flip angle, 23-cm FOV, 256 × 256 matrix, and 1.5-mm section thickness), which was subsequently used for fMRI and SIM coregistration.

During postprocessing of the fMRI results, EPI raw data were low-pass filtered in the spatial frequency domain by using a Hamming filter (20) then reconstructed into individual section-location time-courses. These reconstructed time-course files were checked for patient head motion, and then re-aligned by using the 3Dvolreg spatial registration algorithm of the AFNI software fMRI analysis package (21). A minimum signal intensity threshold was applied to exclude extracranial voxels from further analysis. The time course plots from each unmasked EPI voxel was compared with reference functions with a generalized least-squares fitting algorithm, fitting the observed data voxel-by-voxel to sets of user-defined functions. The fitted functions included a constant (baseline signal intensity level), a ramp (to allow for possible linear signal intensity drift), a temporal smoothing filter (to compensate for the differences in image acquisition times within each 2000-ms TR), and a smoothed boxcar reference function modeling the presumed stimulus responses.

Generation of the EPI SIM

The initial EPIs from each section in the fMRI dataset were thresholded to eliminate signal intensity from regions outside the brain. These images were overlaid on the patient’s anatomic image dataset to create the SIM (Fig 1). Discrepancies due to head motion or misalignment in the SIM were corrected by means of the same linear 6-axis rigid-body translation program used to optimize the coregistration of fMRI maps to the anatomic images.

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

Illustration of SIM map processing. Top row, three noncontiguous coronal EPIs from an fMRI patient dataset show diminished signal intensity from the medial-inferior frontal lobes. Middle row, same sections converted to semitransparent colored masks, which can be superimposed on anatomic images. Bottom row, spatially smoothed version of the masks overlaid on coregistered high-resolution anatomic T1-weighted images. Colorization in the SIM demonstrates regions of adequate EPI signal intensity for detection of the BOLD effect.

Analysis of the SIM

The SIMs were reviewed for evidence of regional signal intensity loss and any regions of signal intensity loss were then classified into one of three types: type I indicated signal intensity absence in the region of the sinuses or skull base; type II, signal intensity absence due to incomplete coverage of the brain with EPIs; and type III, signal intensity absence secondary to susceptibility artifact in regions distant from the sinuses and skull base (Fig 2).

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Types of susceptibility artifact.

A, Type I signal intensity loss in the regions of the sinuses or skull base.

B, Type II signal intensity loss due to incomplete coverage of the brain with the EPI.

C, Type III signal intensity absence in a region of brain distant from the sinuses or skull base. In this case, a cavernous hemangioma with blood products produces the susceptibility effect.

In the cases with type II artifact, the relative amount of section cutoff was calculated as a percentage of the longitudinal dimension of the head that was not covered by the SIM on the axial image on which the cutoff was largest. In the cases with type III artifacts, the size and the location of the signal intensity dropout were tabulated. The cause of the artifact was then determined by examination of pertinent clinical images and reference to the patient’s clinical records. Artifacts caused by surgical clips, metal sutures, burr holes, and other postoperative features were grouped into a category termed surgical site artifact.

The subset of patients with type III artifacts occurring in eloquent brain regions was selected for additional blinded readings. For this purpose, the list of eloquent brain regions included the central sulcus, supplementary motor area, primary auditory cortex, Broca area, and Wernicke area, as anatomically determined by using standard parcellation methods (22). Cases were excluded from the blinded reading if a less-than-10% signal intensity void in the SIM coincided with a region of presumptive eloquent cortex or if no fMRI paradigm to activate the specified eloquent region had been performed. For each patient selected for the blinded study, a list of questions and possible answers was created to test the reader’s ability to recognize the activation with confidence. The questions addressed frontal (Broca area) and posterior (Wernicki area) language dominance, activation of the auditory cortex, and location of the sensorimotor cortex and the supplementary motor area. Questions and answers used in the study are listed in Table 1. Forty questions were posed for the 26 patients that were selected to be included in the blinded readings.

Four readers participated in this study. Readers 1 and 2 were staff neuroradiologists (A.F., H.R.). For these readers, each case was reviewed without the SIM and the questions answered and then reviewed immediately afterward with the SIM and the questions answered again. Readers 1 and 2, who had discussed the study design, had knowledge of the goals of the study but were unfamiliar with the cases chosen. Readers 3 and 4 (D.W., M.H.) were second year neuroradiology fellows. To minimize bias for these readers, the entire set of images minus the SIM was submitted for review before they reviewed the images with the SIM. Readers were told that they were participating in a study to evaluate the effect of disease on fMRI activation. All readers were provided with the patient’s age, sex, reason for the fMRI, a list of the paradigms performed for the fMRI study, a list of questions with answer choices, and a file with all the fMRIs obtained in the patient.

For each reader, the number of times an answer was changed after viewing the SIM was tallied. The type of change was also noted (from specific hemispheric dominance to indeterminate, etc.). Comparisons were made between the more experienced readers (staff) and the less experienced readers (fellows).