Functional Magnetic Resonance imaging

2.2.1.

General overview

In norm al neurological (clinical) use, the m agnetic resonance signal produced by brain tissue com es m ainly from the protons in tissue water, and this signal varies with the local tissue environm ent o f the protons. Placing a body in the large static m agnetic field of an M R I scanner leads to the proton m agnetic m om ents becom ing aligned along the axis of th a t m a g n etic field . By d isp la c in g th e ir alig n m en t th ro u g h e x c ita tio n w ith a radio freq uen cy pulse, and then m easuring the energy prod uced by the longitudinal relaxation o f the m agnetic m om ents back to their original alignm ent, an im age that reflects the tissue env iron m ent o f the protons can be produced. In addition to this longitudinal (T l) relaxation, differences in the m agnetic environm ent o f each spin also lead to differences in the transverse relaxation time (T2) caused by precession o f the spins about the longitudinal axis o f the static m agnetic field. In addition the p resen ce o f param agnetic particles w ithin an object to be im aged (e.g. due to spatial inhom ogeneity) cause an ad ditio nal relax ation , th at to g eth er w ith the T 2 effect is know n as T2* relaxation. In the early 1990s it was realised that variations in T2* could be exploited to m easure cerebral blood flow via the injection of a param agnetic vascular m arker such as gadolinium . T he com bination o f this technique w ith an u ltrafast im age acquisition technique, echoplanar im aging (BPI) (M ansfield, 1977), allow ed dynam ic im aging of the passage o f such a contrast agent through the brain. This lead to the first m easurem ents of changes in cortical perfusion with M RI (B elliveau et al., 1991) , that could be thought of

Neural correlates of selective attention

as loosely analogous to a PET m easurem ent. In this experim ent Belli veau and colleagues com pared the signal returned w hen subjects view ed a flickering visual checkerboard stim ulus w ith a static non-flickering display(B elliveau et aL, 1991). Strong activation rela ted to visual stim ulation w as seen in occip ital cortex. T he nex t step w as the rediscovery that deoxyhaem oglobin is m ore param agnetic than oxyhaem oglobin, m eaning th a t h aem o d y n am ic changes th at alter the rela tiv e p ro p o rtio n o f oxy- to d eo x y ­ haem oglobin will change the m agnetic environm ent and therefore the M R I signal around these cortical vessels. The developm ent o f BPI sequences specifically sensitive to T2* signal (know n as B lood O xygenation Level D ependent or B O LD contrast) allow ed the non-invasive functional im aging o f changes in cortical haem odynam ics related to cerebral activity (K w ong et aL, 1992; O gaw a and al, 1993) This technique is used in the M RI experim ents presented in this thesis.

T he ex act relationship betw een neural activity and BO LD contrast is still un der active investigation (Toga and M azziotta, 1996), but the basic principles have been established using a com bination of optical im aging, single cell electrophysiology and M R I (M alonek and G rinvald, 1996). N eural activity in the brain appears to cause an initial increase in oxygen consum ption, leading to a rise in deoxyhaem oglobin concentration and a fall in the M R signal that can be seen at high field strengths (M enon et aL, 1995). T his brief ‘early d ip ’ is then superseded by a large rise in the signal that is thought to reflect a transient increase in blood flow to the active area. B ecause the flow increase tem porarily o u tw e ig h s th e co n su m p tio n , a larg e tra n s ie n t d e c re a se in d e o x y h a e m o g lo b in concentration occurs, giving rise to a large peak in the M R I signal that is easily seen using BPI at field strengths o f 1.5T and greater. A t the low er field strengths (less than 2 Tesla) used in this thesis, this latter signal is dom inant and the ‘early d ip ’ is usually not seen. N ote that the origin o f the signal depends on an im balance o f flow and m etabolism {see also (Fox et aL, 1988)}. If the resupply o f oxyhaem oglobin alw ays balanced tissue consum ption, no change in the BO LD contrast M R I signal w ould be seen. T echnical issues regarding the origin of the BOLD signal are usually only tangential to the conduct o f cognitive experim ents. H ow ever in this thesis. C hapter 8 contains further discussion and an experim ent explicitly addressing the relative stim ulus dependencies of PBT and fM RI signals, as this is relevant to the rate m odulation technique presented in C hapter 3 . Sim ilarly the relationship betw een neurophysiological findings at a single neuron level (from single cell electrophysiology) and those recorded w ith functional im aging from populations o f several hundred thousand or million neurons is not well understood. The

Neural correlates of selective attention

use of fM R I techniques on aw ake behaving monkeys will begin to address this question (Stefanacci et al., 1998) by allow ing the sim ultaneous recording o f single neuron activity and evoked haem odynam ic activity from cortical areas.

2.2.2.

fMRI experimental design

T he stu dies d escrib ed in this thesis w ere perform ed on the 2T S iem ens V IS IO N (Siem ens, Erlangen) at the W ellcom e D epartm ent o f C ognitive N eurology using B O LD con trast BPI. Each study typically involved scanning betw een six and nine subjects. U nlike PET, fM R I acquisition is typically continuous with each volum e (com prising m ultiple contiguous slices) being acquired every 2-6 seconds. The low er signal to noise ratio, con tinu ou s acquisition and specific artefacts m ake a k now ledge o f po ten tial statistical confounds invaluable in designing effective fM R I experim ents. T he m ost im portant o f these are described briefly here; these issues are described in m ore detail elsew here (Turner eta l., 1998).

H ead m otion is an im portant potential confound for both PET and fM R I experim ents. The e x p erim en ter attem pts to m inim ise head m ovem ent through the use o f soft padded restraints. The dynam ic nature of fM RI data acquisition m eans that not only the position o f the o b ject in the scanner, but also the history of the object w ithin the scanning env iro nm en t w ill affect the signal w ithin a slice. This m eans that even when m otion effects are rem oved through realignm ent o f m ultislice im ages (see section 2.4.1. ), there m ay be residual signal changes related to subject m otion. Typically the repeat tim e (TR) in m ultislice functional im aging studies is of the order of the T l relaxation tim e for brain tissue, so out-of-plane m otion can create a different spin excitation history for different groups o f spins w ithin a slice being im aged. This contam inates the data with m otion correlated signal intensity changes, so an additional autoregressive algorithm is used with fM R I realignm ent to rem ove this com ponent of the signal (Friston et al., 1996b). This represents a practical problem for paradigm s where stim ulus-correlated head m ovem ent occurs, as any signal evoked by task perform ance that is also correlated w ith subject m otion w ill be rem oved by this algorithm . This p articu larly affects sen so rim o to r paradigm s in w hich proxim al lim b m ovem ent occurs. Sim ilarly, paradigm s w here spoken responses are required also produce task-correlated head m o vem ent as w ell as m ore subtle changes in the local m agnetic field hom ogeneity. For this reason experim ents that require spoken responses are inadvisable.

N eural correlates o f selective attention

In addition to m otion-related changes in signal, fM R I im ages contain other sources of noise. M odern scanners rem ain stable over extended scanning periods, so variability in the signal is m ostly physiological in origin (Turner et ah, 1998). A t a typical repetition tim e (TR) o f a few seconds, the dom inant source o f this noise is due to the aliasing of cardiac and respiratory cycles. B ecause cardiac and respiratory cycles are a close, but non-integer m ultiple of the repetition tim e, aliasing o f these frequency com ponents will create an additional artefactual signal o f relatively low frequency. In addition slow changes in blood oxygenation also appear to occur over the course o f an experim ent (B isw al et a l , 1995), giving rise to low frequency drifts and shifts that contam inate the fM R I signal. These low frequency com ponents can be m odelled and rem oved statistically w ith a Fourier series with a relatively low frequency cut off that effects a high pass filter (H o lm es et al., 1997). H o w ev er the p resen ce o f such co n fo u n d s also inform s experim ental design. M ost experim enters choose to design experim ents in such a way that active experim ental conditions alternate in ‘epochs’ o f between thirty and forty seconds. E x perim ental conditions alternate w ith a recurring low level control period, and the objective is to assess the m ean level o f activity over the course o f an epoch in each experim ental condition. The variability in the activity o f the recurring control condition can be used to ch aracterise low frequency drifts and shifts. M ore im portantly, the altern atio n o f experim ental conditions creates a fundam ental frequency fo r the task related activation that is sufficiently high to be clear o f the low frequency region o f the noise spectrum , w ithout being so high as to be attenuated by the sluggish haem odynam ic response.

T he slow rise and fall of the haem odynam ic response that occurs in response to neural activity m eans that the haem odynam ic activity m easured by fM R I is effectively a low pass filtered representation of the neural activity {assuming linearity: see C hapter 8 and (F riston et al., 1998b)}. D uring an experim ental epoch, task perform ance results in successive haem odynam ic responses to repeated neural activity that in effect sum m ate, pro d u cin g a sm ooth m ean level of activity. Inferences can then be m ade ab ou t the variation in that m ean level o f activity w ith som e experim ental m anipulation. As with PE T , all conventional psychological designs can be im plem ented (F rackow iak et al.,

1998). In this thesis, both sim ple 2x2 factorial designs and m ore com plicated param etric factorial designs are used.

Neural correlates of selective attention

M ore recently, advances in experim ental design and analysis have m ade it feasible to attem pt to detect the m odulation o f haem odynam ic activity to individual neural events ( ‘event-related’ fM RI). This is equivalent to tim e locking fM R I acquisition to individual task events, in a m anner analogous to ERP, and seeking evidence for m odulation of the resultant evoked response. This requires that the individual events be separated in tim e during perform ance of a task sufficiently far that the individual haem odynam ic responses to task events can be resolved. The tim ing of individual events can be used to create an event train that is then convolved with a synthetic haem odynam ic response to create a regressor that represents the expected haem odynam ic response to these task events. The fu nctional im aging data can then be exam ined for the presence o f such ev ent-related m o d u latio n . In the su p p o rtin g m aterial to this th esis, an ex p erim en t u sin g this m ethodology is presented (Lum er et a l , 1998). In this experim ent. Dr. E rik L um er and m yself investigated the neural correlates o f spontaneously occurring transition events in bin o cu lar rivalry. By asking subjects to press a button w henever they experien ced a transition betw een different m onocular view s of a rivalrous stim ulus, we w ere able to look for activity specifically correlated in tim e with perceptual transitions. C om pared to a control condition, differential activity tim e locked to these transitions in binocular rivalry was seen only in right frontoparietal structures. In recent w ork we studied how such responses to rivalrous stim uli evoked in visual and extrastriate cortex covaried in space and ov er tim e (Lum er and Rees, 1999). A nalytic techniques for event related fM R I are still evolving, and the interested reader is directed to the relevant literature (Friston et al., 1998a) for further inform ation.

T he o th er fM R I studies described in the thesis used a conventional alternating epoch design. Each scanning run (com prising 80-300 volum es) is also preceded by six or eight ‘d u m m y’ scans to allow for equilibration o f T l relaxation effects. The dum m y scans are discarded prior to analysis. All subjects gave inform ed consent for their participation in the studies, that w ere approved by the N ational H ospital for N eurology & N eurosurgery ethics com m ittee. In addition to the T2* w eighted BOLD contrast volum es, an anatom ical T l w eighted structural im age was also acquired for each subject