Accuracy of Functional Localization in Pre-surgical Function MRI
Mu-Lan Jen1, Islam S. Hassan2, Ping Hou1, Guang Li1, Ashok J. Kumar2, Colen R. Rivka2, and Ho-Ling Liu1

1Department of Imaging Physics, The University of Texas M. D. Anderson Cancer Center, Houston, TX, United States, 2Department of Diagnostic Radiology, The University of Texas M. D. Anderson Cancer Center, Houston, TX, United States

Synopsis

This study evaluates the errors associated with the spatial transformation process by using algorithms commonly applied for clinical pre-surgical fMRI. The images from nine right-handed patients with brain tumors were retrospectively analyzed. Significant differences (P<0.05) were found when comparing results from automated registration (AR) vs. coordinate matching (CM) and AR vs. AR with manual adjustment (AR, adjusted). No statistical significance was found between CM and AR, adjusted. This study established a platform for evaluating the functional localization accuracy in pre-surgical fMRI, and highlighted the necessity of quality control for the AR processing as a clinical routine.

Introduction

Functional MRI (fMRI) is one of the most important pre-surgical brain mapping tools [1], with advantages including non-invasiveness and high spatial resolution. It can provide important spatial information about the location of brain activity in eloquent cortices. However, fMRI data are often acquired using low-resolution EPI and then overlaid on high-resolution volumetric images for surgical navigation. The spatial transformation of detected activation foci to anatomical images is a critical process to maintain the localization accuracy. This study aimed to evaluate the errors associated with this process when using algorithms commonly available and applied for clinical pre-surgical fMRI.

Methods

The pre-surgical MRI exams of nine right-handed patients (2 females, 7 males; 34-68 yr-old) with malignant brain tumors at the fronto-parietal region were retrospectively analyzed. All scans were performed on a 3.0T MR scanner with an 8-channel head coil (GE Healthcare, Waukesha, WI, USA), which consisted of a high resolution 3D T1-weighted structural scan, a 2D T1-weighted structural scan and a gradient-echo EPI functional scan. The 2D T1-weighted imaging was acquired with the exact slice thickness and location matched with the fMRI. We used a block-design experiment constituting of bilateral hand squeeze as an active task alternating with rest. For comparison, all fMRI data was spatially transferred to the 3D T1-weighted images with two algorithms: coordinate matching (CM) using the AFNI software (http://afni.nimh.nih.gov/) and automated registration (AR) using the DynaSuite Neuro 3.0 software (Invivo, Gainesville, FL, USA). For the AR, results were obtained both without and with manual adjustment (AR, adjusted). The functional map (based on correlation analysis) for each patient were overlaid on both of the original EPI volume and the 3D T1-weighted image volume, with proper thresholds to optimize visualization of primary motor area [2]. An experienced neuroradiologist delineated the detected activation blob in the same location on 2D T1-weighted structural images, as those overlays on EPI volume using the Mango software (http://rii.uthscsa.edu/mango/index.html). Then the manually drawn ROIs (serving as the true functional localization) was transferred to the 3D T1-weighted image volume, using the transformation matrix determined by registering the 2D to the 3D T1-weighted image volumes using SPM8 software (http://www.fil.ion.ucl.ac.uk/spm/). The Euclidean distances between the manually drawn activation ROIs and the software generated overlays were determined in the 3D structural image space. The results were then compared by using Wilcoxon matched-pairs signed rank test for each two sets of data.

Results

The Euclidean distance between the centroid of the software generated activation overlay and that of the hand-drawn ROI was found to be 4.7 ± 2.0 mm in CM, 10.1 ± 4.6 mm in AR, and 5.4 ± 2.6 mm in AR, adjusted, respectively. Significant differences were found when comparing results from AR vs. CM and AR vs. AR, adjusted (P<0.05). No statistical significance was found between CM and AR, adjusted.

Discussion and Conclusions

In principle, spatial transformation based on CM alone could suffer from patient motion, whereas simple rigid-body AR could lead to errors due to differences in tissue contrast and extent of lesions between the two images, and probably more significantly, due to the geometric distortions in echo-planar images. This study found that the AR itself could lead to a centroid shift of the activation foci to a distance close to one gyrus, which could be problematic for the surgical planning. The smaller localization error found with CM was a result from good motion control between the functional and anatomical scans, thus cannot guarantee to succeed in all clinical studies. The results from AR, adjusted, showed significant improvements from the AR alone and were comparable to the CM, suggesting the importance of a user-friendly and accurate manual adjustment function in clinical fMRI software. This study established a platform for the evaluation of functional localization accuracy in pre-surgical fMRI, and highlighted the necessity of quality control for the AR processing as a clinical routine.

Acknowledgements

No acknowledgement found.

References

[1] Petrella et al., Radiology, 2006;240(3):793-802;

[2] Kundu B. et al., Neurosurgical Focus, 2013;34(4):E6.

Figures

Figure: Overlays of activation blobs on 3D T1-weighted images. (blue: regions with hand-drawn ROI only; yellow: regions with spatially transformed activation blobs only; red: overlapping regions between the hand-drawn ROI and spatial transformation)

Table: Distance (in mm) between the spatially transformed activation foci and the anatomical location determined based on overlays on echo-planar images.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
3797