Three-dimensional whole-head optical tomography of passive motor evoked responses in the neonate

of 9

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
9 pages
0 downs
Three-dimensional whole-head optical tomography of passive motor evoked responses in the neonate
  See discussions, stats, and author profiles for this publication at: Three-dimensional whole-head opticaltomography of passive motor evoked responsesin the neonate. Neuroimage 30:521–528  Article   in  NeuroImage · May 2006 DOI: 10.1016/j.neuroimage.2005.08.059 · Source: PubMed CITATIONS 96 READS 35 9 authors , including:Adam GibsonUniversity College London 130   PUBLICATIONS   3,282   CITATIONS   SEE PROFILE Nicholas L EverdellUniversity College London 32   PUBLICATIONS   557   CITATIONS   SEE PROFILE Judith MeekUniversity College London Hospitals NHS Fou… 82   PUBLICATIONS   3,528   CITATIONS   SEE PROFILE John Stephen WyattUniversity College London 285   PUBLICATIONS   13,725   CITATIONS   SEE PROFILE All content following this page was uploaded by Nicholas L Everdell on 19 December 2013. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the srcinal documentand are linked to publications on ResearchGate, letting you access and read them immediately.  Three-dimensional whole-head optical tomography of passive motor evoked responses in the neonate A.P. Gibson, a, * T. Austin,  b  N.L. Everdell, a  M. Schweiger, c S.R. Arridge, c J.H. Meek,  b J.S. Wyatt,  b D.T. Delpy, a  and J.C. Hebden a  a   Department of Medical Physics and Bioengineering, University College London, Malet Place Engineering Building, London WC1E 6BT, UK   b  Department of Paediatrics and Child Health, University College London, 5 University Street, London WC1E 6JJ, UK  c  Department of Computer Science, University College London, Malet Place Engineering Building, London WC1E 6BT, UK  Received 13 April 2005; revised 8 July 2005; accepted 30 August 2005Available online 24 October 2005 Optical tomography has been used to reconstruct three-dimensionalimages of the entire neonatal head during motor evoked responses.Data were successfully acquired during passive movement of each armon four out of six infants examined, from which eight sets of bilateralimages of hemodynamic parameters were reconstructed. Six out of theeight images showed the largest change in total hemoglobin in theregion of the contralateral motor cortex. The mean distance betweenthe peak response in the image and the estimated position of thecontralateral motor cortex was 10.8 mm. These results suggest thatoptical tomography may provide an appropriate technique for non-invasive cot-side imaging of brain function. D  2005 Elsevier Inc. All rights reserved. Introduction Optical tomography is a medical imaging technique in whichmeasurements of near-infrared light transmitted across the body areused to obtain images of the optical properties of tissue (Boas et al.,2001a, Schweiger et al., 2003, Gibson et al., 2005a). Unlike the  popular technique of optical  topography , which produces 2Dimages of activated regions on the surface of the brain (Strangmanet al., 2002, Hebden, 2003, Koizumi et al., 2003), optical tomography  produces 3D volumetric images of the whole headand hence can identify changes occurring in deeper tissues. Certaincomponents of tissue have optical properties which are wave-length-dependent, such as the absorption spectra of oxy- anddeoxyhemoglobin. Optical tomography using multiple wave-lengths therefore allows images of the concentration of oxy- anddeoxyhemoglobin ([HbO 2 ], [HHb]) to be generated. Furthermore,an optical imaging system is portable and can be used for safe,continuous monitoring at the bedside.Both preterm and term infants are vulnerable to cerebral injuryin the perinatal period, and such damage can occur deep within the brain (Volpe, 2001). The principal motivation for our work has  been to assess regional cerebral blood volume and oxygenation inthe newborn, both to improve our understanding of the patho- physiology of perinatal brain injury and to identify vulnerableinfants who may benefit from new neuroprotective therapies(Hebden, 2003, Gluckmann et al., 2005). Optical images of the infant cortex have previously been obtained using opt ical top-ography (e.g., Taga et al. (2003), Tsujimoto et al. (2004)), and the adult brain has been imaged by limited volume tomography(Bluestone et al., 2001). However, these techniques provide little information about deeper regions of the brain. We have, therefore,concentrated on reconstructing the full 3D volume by measuringlight which has been t ransmitted across t he whole head. The first work of this nature (Hintz et al., 1999) required scan times of  several hours and used a relatively crude reconstruction techniqueto provide 2D images which did, nevertheless, correlate with thosefrom other imaging methods. More recently, we have generated thefirst 3D optical images of an intraventricular hemorrhage in theneonatal brain which correlated with ultrasound images (Hebden et al., 2002). We have also produced functional images during swingsin blood oxygen and carbon dioxide concentration caused bymodifying the ventilator settings of a severely brain-injured infant (Hebden et al., 2004). These images agreed with the expected  physiological changes.While we have validated images obtained in the laboratory ontissue-equivalent phantoms (Gibson et al., 2003b), clinical validation is necessary before the technique can be widelyaccepted. However, as optical tomography is the only methodwhich can produce images of brain function at the cot-side, there isno ‘‘gold standard’’ imaging technique against which it can becompared. An alternative method by which optical tomography can be validated is to image a known functional change and correlateits localization with known anatomy. In this paper, we present 3Doptical tomography images obtained during passive motor evoked 1053-8119/$ - see front matter   D  2005 Elsevier Inc. All rights reserved.doi:10.1016/j.neuroimage.2005.08.059* Corresponding author. Fax: +44 20 7679 0255.  E-mail address:  (A.P. Gibson). Available online on ScienceDirect ( NeuroImage 30 (2006) 521 – 528  responses induced by raising and lowering an infant’s arm andshow that the reconstructed change in optical properties agreeswith the expected anatomical position of the contralateral motor cortex. Method  Instrumentation To detect light which has been transmitted across the entirehead within a timescale which is clinically acceptable, a powerfullight source and sensitive photon-counting detectors are required.With this in mind, we have built a system called MONSTIR (multi-channel opto-electronic system for time-resolved image recon-struction) which has a fiber laser source and 32 parallel time-resolved single-photon counting detectors (Schmidt et al., 2000). Adual fiber laser (IMRA Inc., USA) provides light pulses of 2-psduration at 780 nm and 815 nm which are coupled sequentially viaan optical fiber switch into one of 32 connectors on the head. Eachfiber consists of a co-axial source fiber surrounded by a detector fiber bundle, and terminates in a connector (an ‘‘optode’’) whichholds the source and detector 10 mm from the scalp. This ensureseven illumination, increases the detection area, reduces the effect of hair beneath the connectors and allows calibration by measurement of back-reflected light (Hebden et al., 2003). Light from the activesource is collected simultaneously by all the other detectors andcoupled into four 8-anode microchannel plate photomultiplier tubes in photon-counting mode. The dynamic range of the detectedlight is reduced by a set of variable optical attenuators, to minimizethe number of detectors which saturate. The arrival times of detected photons are compared to a reference signal from the laser and histograms of photon flight times, or temporal point spreadfunctions (TPSFs) are accumulated. A full set of data consists of 31measurements from each of 32 sources, although in practice,measurements with small source–detector separations wouldsaturate the detectors and are rejected, giving typically 800 TPSFs per image data set. The 32 optodes are held in a semi-rigidthermoplastic helmet which is lined with infrared absorbing foamso that the optodes are optically isolated from each other. Currently,a new helmet is custom-built for each infant (Fig. 1). Light fromthe laser is attenuated to ensure the power is less than 15 mW,which is eyesafe according to the British safety standards(Amendment 2 (2001) to BS EN 60825-1, ‘‘Safety of laser  products’’).  Image reconstruction  Near-infrared light travels through tissue predominantly byscatter, so much of the spatial information in the measured signal islost (a typical mean photon flight time is almost an order of magnitude greater than the time required for the most direct path).The image reconstruction problem is consequently ill posed—achange in optical properties may lead to an arbitrarily large changein the measurements. The solution is highly sensitive to thegeometry of the problem, the positions and coupling of the sourcesand detectors, and errors in the data or the image reconstruction. Toreduce these effects, rather than reconstructing images of theabsolute optical properties during activation, images were obtainedof the difference in the optical properties during activationcompared to a resting state. Evoked responses provide a close toideal situation for this type of imaging as the resting state can beassumed to be identical to the activation state with the exception of a single spatially localized perturbation. We have recently shownthat where reference data are available which closely match dataacquired following a change, linear reconstruction may providesuperior image quality compared to non-linear (Gibson et al.,2005b).The diffusion of light across the head was modeled using thefinite element method (Arridge et al., 2000). This requires a finite element mesh upon whose surface the optodes lie and whichconforms to the shape of the head. The surface of the finite element mesh was generated by warping a known head-like surface so that it passed through the positions of the optodes (Gibson et al., 2003a)which were measured using a 3D digitizing arm (Microscribe 3D,Immersion Co., USA). A finite element mesh was generated fromthis surface using Netgen (Scho¨berl, 1997). The mesh parameters were different for each infant but typically included 20,000elements with quadratic interpolation functions. Minimizing themesh density was not seen as a priority as the Jacobian is pre-computed in linear reconstruction.The information content of the TPSFs was compressed byFourier transforming them and extracting the amplitude  A  and phase  /  at a given frequency (100 MHz was used in this work).This reduces the amount of data which the reconstructionalgorithm has to process and considerably speeds up the imagereconstruction. Prior to image reconstruction, the problem waslinearized by the Rytov approximation (Arridge, 1999) such that  changes in  A  and  /  were related to changes in the absorptioncoefficient   l a   and diffusion coefficient   j  = 1/3( l a   +  l s  V ), where  l s  V is the transport scattering coefficient, by the matrix equation D  A u   ¼  J  A l a J  A j J ul a J uj   D  l a j   ;  ð 1 Þ where  J  x y is the Jacobian of the forward mapping for the change indata type  D  y ¼ D  A u    and change in optical parameter  D  x ¼ D  l a j   . The Jacobian matrix  J ¼  J  A l a J  A j J ul a J uj    was found by solving the diffusion equation with the finite element method,using the forward and adjoint solutions to construct sensitivityfunctions (Arridge and Schweiger, 1995). Fig. 1. Activating the right arm of a neonate wearing a helmet with opticaltomography connectors.  A.P. Gibson et al. / NeuroImage 30 (2006) 521–528 522  Complications arise because  D  y  consists of measurements of  both log amplitude and phase, and  D  x  includes images of bothabsorption coefficient   l a   and reduced scatter coefficient   l s  V . Theseissues were addressed by normalizing  J ,  D  y  and  D  x  as follows:(1)  D  x  and  J  were normalized by dividing by the mean optical parameters to ensure that   l a   and  l s  V  contribute equally to the imagereconstruction. We define a matrix  S ¼  l  a  00  j     and a scaleddata vector x˜ =  S  1  x  such that Eq. (1) becomes D  y ¼ J S D ˜  x x :  ð 2 Þ (2) The magnitude, dimension and error of the log amplitudeand phase data types are different. This was corrected by dividing both by their respective standard deviations. If   r ¼  r  A r u   , where j   x is the standard deviation of data type  x , then Eq. (2) becomes D  y r   ¼  1 r J S D ˜  x x :  ð 3 Þ (3) The image reconstruction is regularized according to the L 2 -norm of the data. This is different for the log amplitude and the phase, implying that each has a different influence on the imagereconstruction. This could be dealt with by applying different regularization parameters for   A  and  / , but instead, the data werenormalized to the sum of the squares of the data, i.e., we define C  diag  C  1 ð Þ  00 diag  C  2 ð Þ    1 ; where  C  1 i  ¼ ~  D  A r  A   2 and  C  2 i ¼ ~  D ur u   2 ; for   i  = 1...  n  where  n  is the number of measurements, and applythis to Eq. (3): C ¼  D  y r   ¼  C r  J S  D ˜  x x :  ð 4 Þ Eq. (4) was inverted following Tikhonov regularization of theMoore–Penrose generalized inverse  D x˜ =  J˜ ( J˜J˜ T +  k I ) C ( D  y /  r ),where  T  denotes the matrix transpose,  I  is the identity matrix and  J˜  =( C  /  r ) JS .(4) The effect of coupling between the optode and the tissuewas included in the reconstruction using the augmented Jacobianmethod described by Boas et al. (2001b).Regularization was performed by adjusting the parameter   k which, in this work, was set at 10% of the maximum singular valueof   JJ T  . After examining a range of   k  from 0.01% to 100%, 10%appeared to give smooth, relatively artefact-free images whichwere quantitatively plausible. This corresponds to a mismatch between data and the model of 10% which is larger than themeasured experimental error of about 1% but includes the error dueto the mismatch between the homogeneous model and the datacollected on the neonatal head. The rescaling of   J ,  D  y  and  D  x affects the optimal  k  in a complicated manner.Separate images of   l a   and  l  s  V  were reconstructed at 780 and 815nm. The absorption images were combined using the Beer– Lambert Law as described by Hueber et al. (2001) to provideimages of   D [HHb] and  D [HbO 2 ] which were summed to provideimages of   D [HbT].  Experimental method  Evoked response studies were performed on six very preterminfants at about 1 month of age (Table 1). The infants were selected for study because they were of appropriate head circumference andclinically stable. The median (range) postmenstrual age at birth was28 weeks 5 days (28w 2d to 32w 1d), and the median (range)corrected age when imaged was 34 weeks 5 days (33w 6d to 36w5d). Two pairs of the babies were twins. Two of the babies hadintraventricular hemorrhages at birth, one of which had resolved bythe time of the study leaving a dilated ventricle. Ethical permissionfor the study was obtained from the local ethics committee, andinformed consent was obtained from the parents prior to the study.A helmet was custom-built for each baby which could hold upto 32 optodes. For the smallest babies, there was only room on thehead to support 28 opt odes while maintaining adequate opticalisolation between each (Fig. 1). To detect light which has been transmitted across the wholehead with adequate signal-to-noise ratio, it was necessary to sumthe signal from each source for 10 s. Light is detected on allchannels in parallel, so a single image takes about 11 min toacquire (10 s  32 sources, plus data download and switching time between sources). This is too long for imaging evoked responseswhich reduce in amplitude due to habituation after about 30 s evenif the activation task is maintained. For this study, the effective Table 1Summary of images obtained from the 6 babies examinedCorrectedage at studyCranial ultrasoundappearanceDataobtained?Distance from peak [HbT] tocontralateral motor cortexPeak  D [HbT]( A M)Baby 1 33 weeks, 6 days Bilateral intraventricular hemorrhage NO – – Baby 2 33 weeks, 5 days Normal YES L 5 mm 26R No localized change – Baby 3 36 weeks, 5 days Normal YES L 18 mmR 14.8 mm 15Baby 4 36 weeks, 5 days Normal NO – – Baby 5 34 weeks, 5 days Normal YES L 10.8 mm 10R 5.2 mm 36Baby 6 34 weeks, 5 days Left intraventricular hemorrhageYES L No localized change – R 11.5 mm 25Mean 10.8 mm 25Babies 3 and 4, and babies 5 and 6, are twins. Images acquired during left passive motor activation on babies 2 and 3 are shown in Figs. 4 and 5, respectively.  A.P. Gibson et al. / NeuroImage 30 (2006) 521–528  523  acquisition time was reduced to 10 s by activating a single sourceand recording resting baseline activity for 10 s, then inducingactivation by passive movement of an arm for 15 s, with activationdata acquired using the same source during the last 10 s (Fig. 2).This was repeated for up to 12 different sources; after this, the babytended to become restless. In practice, this provided about 320measurements (source–detector pairs) per image.The activation exercise involved passively raising and loweringthe left or right arm. This activation can be performed on a sleeping baby, stimulating activity which is spatially localized to the sensoryand motor cortices, and providing good spatial separation betweenleft and right cortical activity. In total, a full experiment involving12 sources with bilateral activation including calibration measure-ments took about 2 h, so it could be conducted between feedingtimes when the baby was most passive. Results  Data quality Data were successf ully acquired from four of the six infantsexamined (see Table 1). The two infants on whom data were not successfully acquired were the first infant we examined, when theexperimental procedure was still being developed, and a later studywhich was affected by movement artefact. Both these studies wereabandoned, and images were not reconstructed.Light leaking from a source around the head to a detector due toinsufficient contact between the head and the helmet was arecurring problem. Babies were positioned with their head raised tominimize the effect of gastro-esophageal reflux. This reducesirritation and therefore movement but  increases the likelihood that the baby will slip out of the helmet. Fig. 3 shows temporal point spread functions (TPSFs) acquired for a single source–detector combination during activation (top) and at rest (bottom). TheTPSFs are smooth and contain about 400,000 photons. The TPSFacquired during activation can be seen to be preceded by a ‘‘pre- peak’’ (Hillman et al., 2000). This is due to light leaking betweenoptodes through air and therefore arriving before the diffuse light which has passed through the head. Such a pre-peak has a greater effect on the amplitude and mean photon flight time than that of theactivation and will therefore lead to artefacts in the image.Examination of the time-domain data allows erroneous TPSFssuch as these to be identified and eliminated from the imagereconstruction. The remaining data types were then examinedfurther by plotting them against optode separation, and anyoutliers, determined by their deviation from an approximatestraight line fit, were excluded. Typically between 30 and 60measurements were rejected out of a total of 320, usually becauseof light leakage or intermittent instability in one of the detector channels.  Images of absorption and scatter  Images of bilateral evoked responses were successfullyreconstructed from the remaining four infants. Six of the eight sets of images showed an increase in optical absorption in theregion of the contralateral motor cortex at both wavelengths. Themean ( T standard deviation) increase in absorption in the 6successfully reconstructed images was (0.0129  T  0.0016) mm -1 at 780 nm and (0.0123  T  0.0011) mm -1 at 815 nm. This represents alocalized increase in absorption of approximately 29% at 780 nm at which deoxyhemoglobin absorbs most strongly, and 23% at 815nm where oxyhemoglobin is the strongest absorber. By compa-rison, the maximum change in scatter was 5% (the mean was 2%).The scatter images showed little physiological change andconsisted primarily of measurement noise and cross-talk from thelarger change in absorption.  Images of functional change The reconstructed images of absorption at 780 and 815 nmwere converted into images of absolute change in oxy- and deoxy-and total hemoglobin ( D [HbO 2 ],  D [HHb] and  D [HbT], respec-tively) as described in Section 2.2. The position of the representa-tion of the arm in the motor cortex was estimated from knowledgeof the anatomy of the neonatal brain and, in the six out of eight examples which demonstrated correctly localized images of absorption, the peak   D [HbT] was in the region of the contralateralmotor cortex. The mean distance from the peak change in [HbT] tothe estimated position of the motor cortex was 10.8 mm (see Table1). Fig. 4 shows a series of sagittal slices of   D [HbT] obtainedduring activation of the left arm. The peak change in the imageoccurs 18.0 mm from the estimated position of the right motor cortex (denoted by a cross), which was the largest error observed.Fig. 5 shows similar images of  D [HbO 2 ], D [HHb] and D [HbT] in adifferent individual. Fig. 2. Schematic illustration of the activation paradigm. The horizontalgrey bar shows the period of activation.Fig. 3. Temporal point spread functions acquired on an infant. The topimage shows data acquired during stimulation, the bottom image is thecorresponding reference data. The pre-peak visible on the top image is dueto light escaping from the head and will overwhelm any changes in theTPSF due to the stimulation. Effects such as these can be eliminated byexamination of time-domain data.  A.P. Gibson et al. / NeuroImage 30 (2006) 521–528 524
Related Search
Similar documents
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks