Análisis del sector económico en Risaralda

When the first stimulus in each block was added to the model (Analysis 2), the effect o f reading relative to fixation for the event-related model only was no longer detected even when lowering the statistical threshold to /?<0.001 (uncorrected for multiple comparisons) (see Table 7.1). In other words, the event-related model no longer explained changes in BOLD signal that were not accounted for by the epoch model. In contrast, when the last stimulus in each block was modelled (Analysis 3), the effect o f reading relative to fixation for the event-related model only was still significant, with Z- scores slightly higher than in Analysis 1 (see Table 7.1).

Occipital

Reading>Fixation (Event-Related Regressor only)

Analysis 1 Analysis 2 Analysis 3

Left Fusiform -18--102 2 5.9 n.s. 6.0

Right Fusiform 18 -88 -12 7.0 n.s. 7.7

Left Lingual -8 -96 -2 6.2 n.s. 7.1

Right Middle Occipital 32 -88 -2 5.6 n.s. 6.8

Temporal

Left Posterior Inferior Temporal -36 -66 -18 6.5 n.s. 7.3

Right Posterior Inferior Temporal 38 -66 -18 5.6 n.s. 7.1

Left Superior Temporal -54 -50 10 6.5 n.s. 7.0

Right Superior Temporal 62 ^ 8 10 5.8 n.s. 5.6

Right Middle Temporal 64 -16 -10 6.3 n.s. 7.0

Parietal

Right Angular gyrus 24 -80 42 6.0 n.s. 6.7

Frontal

Right Inferior Frontal 44 12 28 6.3 n.s. 6.9

Left Inferior Frontal -36 6 26 6.1 n.s. 6.5

T able 7.1 E ffects o f reading relative to fixation th at w ere exp lain ed b y th e even t-related b ut n o t b y the epoch m odel (A n alysis 1). T h ese effects w ere no lo n g er sign ifican t (n.s.) w h en the d ifferen ce in the o n set o f the epoch and the event-related m od el w as d iscou n ted (A n alysis 2). In con trast, they

7.4 Discussion

In the present study, the relative sensitivity o f standard epoch and event-related analyses was investigated using a blocked design reading paradigm. In order to avoid anecdotal comparison o f the contrast-specific statistics, which may be confounded by differential efficiency o f response estimation for epoch and event-related models, a mixed model was used in which trials were characterised as both epochs and single events (Analysis 1). This allowed direct estimation o f the additional variance in the BOLD signal that was explained by either model using the extra sum o f squares principle. The results show that, in critical language regions, a significant amount o f variance in the BOLD signal was accounted for by the event-related regressor that was not modelled by the epoch-related regressor. Although activation in these regions was reliably detected using an epoch model (see Chapter 6; Mechelli et a l, 2000), the present study shows that an event-related approach may characterise the observed response with greater finesse, thereby maximising sensitivity.

In order to determine whether the advantage o f the event-related model lay in the early onset or the late offset o f the modelled response, the first and the last stimuli in each epoch were also modelled (Analysis 2 & 3 respectively). When the early onset of the event-related model was discounted (Analysis 2), the event-related model no longer showed an advantage. In contrast, when the late offset o f the event-related model was discounted (Analysis 3), the event-related model retained its ability to explain changes in BOLD signal that were not accounted for by the epoch model. This indicates that the advantage o f the event-related model lies in its early onset rather than its late offset. One possibility for this asymmetry is that adaptation renders the stationary linear model for

evoked responses less appropriate at the end o f a block. It should be noted that the advantage o f the event related model was still evident when temporal derivatives (which model slight onset and offset differences) were included in the epoch model. This suggests that the advantage o f the event-related model was not simply due to a difference in latency but also to a difference in the shape o f the haemodynamic response.

Although the shape o f the overall haemodynamic response may be affected by nonlinearities in the BOLD signal, especially for short SOAs or long epoch lengths (see Chapter 4), it is important to note that nonlinearities in the amplitude o f the haemodynamic response as a function o f stimulus rate cannot explain the advantage of event-related over epoch-related analyses. This is because each stimulus rate was modelled separately, allowing for any nonlinear relationship between rate of presentation and response (see Friston et a l, 1998 for a full discussion o f this in relation to generalized convolution models).

7.5 C onclusion

The use o f a mixed model, in which trials were modelled as both single events and epochs, allowed the estimation o f the variance in the BOLD signal that was explained by either the event-related or epoch regressors having discounted the effect o f the other. The results indicate that, in a number o f critical language regions, an event-related model may characterise the form (particularly the onset) o f the observed haemodynamic response better than an epoch-related model, even in the context o f blocked design fMRI. Better model fit will reduce error variance thereby increasing the sensitivity of the analysis.

Chapter 8

Neuroimaging studies of word and pseudoword reading: consistencies,

inconsistencies and limitations

8.1 Introduction

This chapter is concerned with why functional imaging studies in different groups o f subjects sometimes result in very inconsistent patterns o f activation. The paradigm that is explored is one where reading words is contrasted to reading pseudowords - a paradigm which is often chosen to test different cognitive models o f reading. Below I briefly describe the models that have motivated a number o f functional imaging studies. Then I discuss the ambiguous results and suggest hypotheses to explain the consistencies and inconsistencies in the literature. Finally I present data from an fMRI study that investigates the effect o f word type and its interaction with stimulus duration.