2.3.1. Behavioural receiver operating characteristic (ROC) analysis
Hit and false alarm pairs for each level of confidence for faces and for scenes were plotted, and ROC curves were fitted to these points using a sums-of-squares search algorithm (Yonelinas et al., 1998), which provided a means of estimating the contributions of recollection and familiarity to recognition memory (see Section
1.4.1.3).
2.3.2. Whole-brain fM RI data analysis
The data were submitted to a (random effects) general linear model, with one predictor that was convolved with a standard model of the haemodynamic response function (HRF) for each event-type. Separate regressors were modelled for each event-type at encoding. These were determined by stimulus (face/scene), and subsequent memory response (high, medium and low confidence old or new). Levels of response accuracy and the distributions of confidence responses guided the formation of 4 encoding regressors for each stimulus category. These were: (a) subsequent miss (sM) - study items that subsequently received high, medium or low confidence new responses, (b) subsequent probably old (sPO) - study items that subsequently received medium and low confidence old responses, (c) subsequent sure old (sSO) - study items that subsequently received high confidence old responses and (d) a regressor for study items that received no response at test.
Retrieval event-types were determined by stimulus type, item status (old/new) and participant response (high, medium and low confidence old or new). There were 5 retrieval regressors for each stimulus type: (a) correct rejection (CR) - high, medium
and low confidence new responses to new items, (b) miss (M) - high, medium and low confidence new responses to old items, (c) probably old (PO) - medium and low confidence old responses to old items, (d) sure old (SO) - high confidence old responses to old items, and finally (e) a regressor of no interest that comprised false alarms (all old responses to new items) and trials where participants failed to make a response.
Parameter estimates relating the height of the HRF response to each event-type were calculated, on a voxel by voxel basis, via a multiple linear regression of the response time-course, to create one beta image for each event-type per run, per participant. Individual runs were then concatenated for each participant in a fixed effects analysis using FEAT. The subsequent parameter estimate images were then combined in a higher-level (group) FLAME analysis (FMRIB's Local Analysis of Mixed Effects: Beckmann, Jenkinson, & Smith, 2003; Woolrich, Behrens, Beckmann, Jenkinson, & Smith, 2004). To examine significant encoding and retrieval memory effects for the two stimuli at the whole-brain level, categorical contrasts between memory regressors were performed separately for faces and scenes. For encoding, whole-brain contrasts were between the face/scene subsequent sure old and miss regressors (scene sSO > scene sM; face sSO > face sM), whereas at retrieval, the contrasts were between face/scene sure old, correct rejection and miss regressors (scene SO > scene CR; scene CR > scene SO; scene SO > scene M; face SO > face CR; face CR > face SO; face SO > face M). FEAT group (gaussianised) t-statistics were then converted to z- statistics and thresholded at p<0.001, uncorrected for multiple comparisons; significant activations involving contiguous clusters of at least 9 voxels are reported. A voxelwise approach was employed as there is evidence that functional differences between faces and scenes can occur within MTL regions (e.g. anterior vs. posterior hippocampus for faces and scenes, respectively; Lee et al., 2008) and because it was expected that significant effects may be evident in smaller MTL regions (e.g. perirhinal cortex). A probability of p<0.001 with an extent threshold of >9 voxels is equivalent to a mapwise false-positive rate for the MTL (encompassing the hippocampus, parahippocampal gyrus and perirhinal cortex) of p<0.05 (estimated using the Monte Carlo procedure implemented in the AlphaSim program in Analysis of Functional Neuroimages (AFNI)). The locations of significant effects were
identified using the Harvard-Oxford sub-cortical structural atlas in FSLView; co ordinates (x, y, z) of significant effects are reported in MNI space (see Section 2.2.5).
2,3.3. Functional region o f interest (fROI) JM R I analysis
Due to the specific aims of this experiment a functional region of interest (fROI) analysis was conducted. This involved identifying stimulus-specific voxels within different subregions of the MTL and investigating memory effects for faces and scenes within each of these. This analysis strategy was driven by observations of stimulus-specific processing within the MTL (Awipi & Davachi, 2008; Barense, Henson et al., 2010; Lee, Bandelow et al., 2006; Lee et al., 2008; Litman et al., 2009), and by accounts which suggest that areas of the MTL that process different types of complex visual stimuli will also support memory for these (Graham et al., 2010; Lee, Barense et al., 2005; see also Bussey & Saksida, 2005; Saksida & Bussey, 2010). To create unbiased stimulus-specific fROIs, group-level contrasts2 were performed between the regressors for novel faces and scenes3 (CR faces > CR scenes and the reverse). These orthogonal correct rejection contrasts between faces and scenes were undertaken within three anatomically-defined MTL ROIs in MNI space (Fig. 2.3); perirhinal cortex, hippocampus and parahippocampal gyrus. The perirhinal cortex was defined using a probabilistic map taken from Devlin and Price (2007) (available at http://joedevlin.psychol.ucl.ac.uk/perirhinal.php), which was restricted to an area that comprised a > 50% likelihood of being the perirhinal cortex in their participants (N = 12). The hippocampus and parahippocampal gyrus were defined using the Automated Anatomical Labelling (AAL) brain atlas (Tzourio-Mazoyer et al., 2002). Any voxels from the parahippocampal gyrus or hippocampal masks that overlapped with the probabilistic map of the perirhinal cortex were removed. The resulting FEAT t- statistics were converted to z-statistics and a liberal (uncorrected) voxel threshold of p<0.025 was applied to the data to ensure stimulus-specific voxels associated with the
2 As there is high consistency across experiments for extrastriate fROIs derived at the group level (Duncan, Pattamadilok, Knierim, & Devlin, 2009).
3 Novel items were used to derive these fROIs as one can then make inferences about memory effects by observing the differences in activity between hit and miss categories, which in some cases may be preferable due to incidental memory encoding activity associated with correct rejections (Stark & Okado, 2003).
Figure 2.3: L eft sagittal (left), coronal (middle) and right sagittal (right) views o f the hippocampal (green), parahippocam pal gyrus (pink) and perirh inal cortex (blue) masks used to create fR O Is in all 3 experim ents in this thesis. Im ages are rendered on a M N I-152 T l 2mm standard brain.
task were identified within each anatomical region. Data are reported for functional ROIs that comprised 10 or more activated voxels. This produced left and right hemisphere fROIs for faces (Fig. 2.4) within anterior hippocampus (left peak: -22, -8, -24, z = 5.31, 237 voxels; right peak: 18, -4, -16, z = 4.99, 289 voxels), anterior parahippocampal gyrus (left peak: -20, -10, -24, z = 5.00, 247 voxels; right peak: 20, 0, -20, z = 5.43, 349 voxels) and perirhinal cortex (left peak: -32, -4, -34, z = 3.77, 74 voxels; right peak: 24, 0, -28, z - 3.89, 248 voxels), and for scenes (Fig. 2.4) within posterior hippocampus (left peak: -18, -34, 4, z = 3.75, 125 voxels; right peak: 20, - 32, 2, z = 3.74, 134 voxels) and posterior parahippocampal gyrus (left peak: -24, -34, -8, z = 3.34, 107 voxels; right peak: 16, -40, -8, z = 4.37, 177 voxels).
FROI analyses were conducted on the data from the encoding and retrieval phases, which are summarised in separate sections. Using Featquery from the FSL toolkit, the following mean parameter estimate values were extracted from each of the stimulus- specific fROIs for the sSO, sPO and sM face and scene regressors from encoding, and the CR4 SO, PO and M face and scene regressors from retrieval. Parameter estimate values were scaled by the height of the effective regressor and mean voxel intensity in order to convert them into percent signal change.
4 For demonstration purposes values for CR have been plotted alongside SO, PO and M in the results. To ensure that the contrasts used to identify stimulus-specific voxels were orthogonal to the investigation o f memory-related activity, CRs were not included in the analyses.
Individual percent signal change values were entered into a 2*2*3 ANOVA with factors of ‘hemisphere’5 (right vs. left), ‘stimulus’ (face vs. scene) and ‘memory’ (encoding: sSO, sPO and sM; retrieval: SO, PO and M) for each fROI. These were conducted separately for the data from encoding and retrieval phases. If there was no significant hemisphere*memory or hemisphere*stimuli*memory interaction, effects were collapsed across hemisphere for the purposes of follow-up analyses. Memory effects were also considered separately for each stimulus type within each fROI (and hemisphere where necessary), by conducting one-way ANOVAs with levels of sSO, sPO and sM (encoding) and SO, PO and M (retrieval). When a significant main effect or linear trend (as indicated by within-subject contrasts) was obtained in these one way ANOVAs, subsequent planned pairwise comparisons were conducted to identify the reason for the reliable effects.
Figure 2.4: Left sagittal (left), coronal (middle) and right sagittal (right) views o f fun ctional ROIs located within the hippocampus (green), parahippocam pal gyrus (pink) and perirhinal cortex (blue) derived fro m the (A) CR fa c e > CR scene and the (B) CR scene > CR fa c e contrasts, rendered on a M N I-152 T1 2mm standard brain.
h em isp here was included as a factor because o f reports o f right-lateralised effects within the MTL for different types o f complex visual stimuli (Awipi & Davachi, 2008; Lee, Bandelow et al., 2006; O'Neil et al., 2009; Rudebeck & Lee, 2010).