3T related problems regarding presurgical mapping of eloquent brain areas

Purpose: fMRI profits in several ways from an increasing field strengths. The BOLD-to-noise ratio as well as the detected cluster sizes are almost doubled when moving from 1.5T to 3T. The detection of higher cognitive functions becomes easier due to an increase in sensitivity. The signal gain can also be used to compensate the signal losses accompanied with parallel imaging techniques as SENSE, when high spatial resolution and / or time resolution are strived for.

However when it comes to the clinical applications high field fMRI might also have some downsides: Increased susceptibility artefacts and image distortions leads to difficulties for intra-modal coregistration between functional and reference data sets. In addition one of the biggest advantage of high field fMRI, the increased sensitivity for bold changes, can become a disadvantage when the increased size of activated clusters lead to cluster fusion when spatial smoothing is applied. In the following study we focused on the latter point, and investigated if the Threshold Clustering Algorithm as described by: ... or Independent Component Analysis ... can solve the problem of excessive cluster fusion.

Methods: 17 patients, (8 males, 9 females) with different central lesions underwent fMRI before neurosurgery. A self paced finger tapping or fist clenching paradigm was used depending on the ability of the patient. Each run of both paradigms consisted of six 30sec on/ off blocks. In all subjects we test both hands. All exams were performed on clinical 3T scanner (Philips Gyroscan Intera) using a Single Shot Gradient Echo sequence. The scan parameters were as follows: Sequence: TE/TR/flip: 35/3000/90°, FOV: 230×230 mm2, voxel size: 3.59×3.59×3.59 mm3. The preprocessing and the analysis was performed using the FMRIB's Software Library (FSL). The preprocessing included motion correction using McFLIRT [Jenkinson 2002], non-brain removal using BET[Smith 2002], spatial smoothing using a Gaussian kernel of FWHM 3.59mm, global (volumetric) multiplicative mean intensity renormalization and temporal highpass filtering (1/100s). The time series where than analyzed using ICA [Hyvärinen 1999], [Beckmann 2002] and FILM (FMRIB's Improved Linear Model) with local autocorrelation correction. The latter includes the Threshold Clustering Algorithm. For the ICA-analysis a threshold of 0,5 was used. The FILM analysis was thresholded using different z-cluster level (z>2.3, 3.3, 4.3, 5.3, 6.3, 7.3, 8.3) and an overall significance threshold of p<0.01(corrected). The different z-cluster levels were choosen in order to test for the optimum z-cluster threshold leading to stable cluster sizes.

Results: Graphing the average cluster size in PM1 versus the z-cluster threshold used showed stable results above a z-threshold of 6,3. A further increase of the z-cluster size did not change relevantly the size of the detected clusters. Using this 'optimal' z-cluster threshold we were able to detect PM1 and cerebellar activation in all cases (both sides and in all volunteers). We failed only one time to detect SMA activation when this method was used. Using Independent Component Analysis found the same success rate except one more failure in detecting SMA activation. The detected cluster size were comparable between both methods. PM1 (ipsilateral) 34±8 vs. 30±22, PM1 (contralateral): 42±12 vs. 38±13 and SMA: 10±12 vs. 8±15 if z-clustering was applied or for the ICA-analysis respectively.

Discussion: Both methods: FMRIB's Improved Linear Model with local autocorrelation correction in combination with the threshold clustering algorithm and Independent Component Analysis provide comparable and stable results. Both methods do not rely on excessive smoothing and are able to solve the problem of excessive cluster fusion which is crucial if high field fMRI is applied for presurgical mapping of eloquent brain functions.

References: [Jenkinson 2002] M. Jenkinson and P. Bannister and M. Brady and S. Smith. Improved optimisation for the robust and accurate linear registration and motion correction of brain images. NeuroImage 17:2(825–841) 2002.

[Smith 2002] S. Smith. Fast Robust Automated Brain Extraction. Human Brain Mapping 17:3 (143–55), 2002

[Hyvärinen 1999] A. Hyvärinen. Fast and Robust Fixed-Point Algorithms for Independent Component Analysis. IEEE Transactions on Neural Networks 10(3):626–634, 1999.

[Beckmann 2002] C.F. Beckmann and S.M. Smith. Probabilistic Independent Component Analysis for FMRI. Technical Report TR02CB1, FMRIB Analysis Group, 2002.