The visually and acoustically presented words evoked three responses named as PM1, PM2 and PM3, respectively. The results of magnetic source imaging ...
Time Course and Neural Network for Comparing Written and Spoken Words: A MEG and DTI Study Lu Meng1,2,*, Jing Xiang1, Douglas Rose1, Rupesh Kotecha1, Jennifer Vannest3, Anna Byars3, and Ton Degrauw3 1
MEG Center, Cincinnati Children’s Hospital Medical Center, 3333 Burnet Avenue, Cincinnati, OH, 45220, USA Key Laboratory of Medical Image Computing of Ministry of Education, Northeast University, Shenyang, 110004, China 3 Department of Neurology, Cincinnati Children’s Hospital Medical Center, 3333 Burnet Avenue, Cincinnati, OH, 45220, USA 2
Abstract— The present study was designed to investigate the spatiotemporal signature as well as the neural network of brain activation associated with word recognition. Twenty participants were studied with a whole head magnetoencephalography (MEG) system and a 3T magnetic resonance imaging (MRI) scanner. The word stimuli consisted of 100 matched words (the visually and acoustically presented words were the same) and 20 mismatched words (the visually and acoustically presented words were different). The time course of brain activation was analyzed with a virtual sensor technique at source levels. The spatial location and connection were estimated with wavelet-based beamformer and diffusion tensor imaging (DTI) tractography. To our knowledge, this is the first study focusing on the time course and neural substrates involved in comparing spoken and written words with both MEG and DTI. The visually and acoustically presented words evoked three responses named as PM1, PM2 and PM3, respectively. The results of magnetic source imaging showed that the PM1 was localized to the primary auditory and visual cortices. The PM2 and PM3 were localized to Wernicke's area and Broca's area. Interestingly, virtual sensor waveforms showed a clear response between PM1 and PM2 from the left temporaloccipital junction. This component was consistently identified around 162 ± 8.9 ms (M160) in 20 participants (100%, 20/20). MEG source-guided DTI tractography revealed neural fibers in this junction area. The results suggested that the M160 was one part of the neural network of auditory and visual word recognition. The results of the present study yield highly convincing evidence that the left temporal-occipital junction plays an important role in comparing visually and acoustically presented words. Keywords— Word recognition, Magnetoencephalography, Virtual sensor, Auditory/Visual evoked magnetic fields, Diffusion tensor imaging.
I. INTRODUCTION Magnetoencephalography (MEG) is an imaging technique used to measure the magnetic fields produced by electrical activity in the brain [1]. Previous reports have shown that MEG reliably records the morphology, latency and amplitude of neuronal activities in real time with
excellent spatial resolution [2,3]. Recent advances in signal processing technology have also made it possible to localize specific volumes brain activation and compute virtual sensors in the brain. Wavelet based beamformer is one of the newly developed methods for analysis of MEG data for this purpose [4]. In comparison to the conventional dipole modeling, wavelet-based beamformer can scan the entire brain for possible sources without making any assumption of the number of sources. Diffusion tensor imaging (DTI) can provide information about white matter fiber tracts in the brain. DTI tractography is a procedure to demonstrate neural tracts by acquiring DTI in at least six non-collinear directions. The result is a 3D network formed by long and short connections among different cortical and subcortical regions, and the network discloses the neuronal connections within biological tissues. The combination of MEG with DTI has the possibility to track neural activation and propagation [5]. To our knowledge, the time course and neural network involved in comparing spoken and written words with both MEG and DTI have not been reported. The objective of the present study was to investigate the neural network associated with word processing with a specific focus on the interaction of the visual, auditory and language systems. The spatial location of word processing in the brain was investigated using wavelet-based beamformer. The time course and connectivity of word processing in the brain was investigated with virtual sensor technique and DTI tractography.
II. MATERIALS AND METHODS A. Subjects and Language Stimuli Twenty healthy, native English-speaking adults were recruited for the present study. In the twenty volunteers, all participants (age: 19-49, mean age: 30 years; 10 female and 10 male) qualified for the present study. A written informed consent approved by the Institutional Review Board (IRB) at Cincinnati Children’s Hospital Medical Center (CCHMC),
S. Supek and A. Sušac (Eds.): Advances in Biomagnetism – BIOMAG2010, IFMBE Proceedings 28, pp. 338–341, 2010. www.springerlink.com
Time Course and Neural Network for Comparing Written and Spoken Words: A MEG and DTI Study
was obtained from each participant prior to testing. All participants were right-handed as measured by a questionnaire based on the Edinburgh Handedness Inventory. The stimuli consisted of 120 words. The words were presented both visually on a backlit screen positioned at a comfortable position in front of the participants and auditorily through earphones. One hundred stimuli consisted of matched words, which meant that the seen and heard word were identical; twenty stimuli consisted of mismatched words, which meant that the seen word was different from the heard word. All the auditory and visual words were presented simultaneously. If the words did not match, the subjects were asked to press a button; no response was required when the visual and auditory stimuli matched. The onset of the visual and auditory presentation was aligned. Word presentation and response recording were accomplished using BrainX software [1]. B. Data Acquisition A 275-channel whole cortex CTF OMEGA MEG system was used for recordings (VSM Ltd., Port Coquitlam, Canada). The MEG measurements were performed in a magnetic shielded room (MSR) with a system white noise level below 10 fT/√HZ. All subjects were supine during the whole data acquisition procedure, their left and right arms rested on either side. The location of the subject’s head relative to the sensor array was measured using three small coils affixed to the nasion and the left and right preauricular points. The three coils were used to measure the position of the sensor array with respect to the nasion-ear coordinate system by means of continuous head localization, which tracks head motion during the entire procedure. MRI scans were recorded by research technologists for all participants with a 3T MRI scanner (Siemens, Erlangen, Germany). T1-weighted MR images were recorded with the resolution of 256×256 and 128 slices for each serial. Aiming at the co-registration of MEG data and MRI data, we placed three fiducial points at the same locations as the positions of the three coils used in the MEG recordings. Consequently, we used them as landmarks to co-register these two data sets. All of the DTI images were acquired on a 3T Siemens Trio MR imaging scanner (Siemens, Erlangen, Germany). A 46-section, diffusion-weighted, spin-echo echo-planar imaging scan was acquired in the axial plane with the following parameters: TR/TE=6000/87 ms; FOV=25.6×25.6 cm; matrix=128×128; section thickness=2mm; b-value=1000 s/mm2; and 4 repetitions. Reference T2-weighted images (b=0) were also acquired. The duration of the DTI sequence was 5 minutes 48 seconds.
339
C. Data Analysis MEG data were averaged and then filtered with a bandpass filter of 1-200 Hz. Neuromagnetic responses (or deflections) were then identified with the averaged data using DataEditor (VSM MedTech Ltd., Port Coquitlam, BC, Canada). Subsequently, the latencies and the amplitudes of each recognizable response were measured using DataEditor. Magnetic sources were localized using wavelet-based beamformer. Wavelet-based beamformer consisted of creating a three-dimensional lattice of spatial filters at 2.5 mm resolution throughout the brain, where the output of each spatial filter was an estimate of the source strength of a single source at that location. Thus, the result of the estimation of the sources was a volumetric image. The intensity of each voxel was the strength of the brain activation at the designated location. Beamformer images were fused into MRI using Magnetic Source Locator (MSL) software based on the three matched fiducial points. Virtual sensors were also computed with wavelet-based beamformer. In this study, the virtual sensors were placed within the functional areas that were localized by waveletbased beamformer at the voxel locations that showed peaks in functional activation. MEG virtual sensor waveforms at these locations were used to analyze the time course of neuromagnetic activity. The latency and amplitude of the response peak of the virtual sensor waveform were measured with MEG Processor. DTI tractography was analyzed and computed using Riemannian framework. Pennec X and colleagues developed a Riemannian framework for the space of 2nd order tensors and recognized the whole brain white matter as a Riemannian manifold. MEG results were coregistered with DTI scans to provide Regions of Interest (ROI) for evaluating tractography between those regions. Connections were found between sites of high coherence in each subject via white matter fiber tracts. We used the software toolkit named DTI track to perform the whole procedure.
III. RESULTS The waveforms corresponding to the matched words showed at least three clear reproducible responses. The three responses were named as PM1, PM2, PM3, respectively (See Fig.1). These three responses’ latencies of averaged MEG physical waveform components were 104.10±5.20 ms, 198.80±6.00 ms, and 302.10±10.01 ms, respectively. Differences were tested for significance using a paired t-test (p
