2.6. Retrieval
2.7.3. MTL activation and recognition confidence
During the retrieval phase, participants were required to make memory judgements to old/new test items on a 6-point confidence scale (1 = high confidence new to 6 = high confidence old), which were used to subcategorise encoding and retrieval items into one of three bins; sure old (sSO/SO), probably old (sPO/PO) and miss (sM/M). In this study all significant memory effects at encoding and retrieval comprised increases in activity for old items according to confidence in the old decision. Furthermore, planned pairwise comparisons between memory confidence bins revealed evidence of a face-specific recollection effect at retrieval, as indexed by an increase in brain activity for sure old, relative to probably old and forgotten items (sSO/SO > sPO/PO = sM/M) within the anterior hippocampus (Daselaar et al., 2006; Yonelinas et al., 2005). In contrast, pairwise comparisons failed to reveal any reliable overall hit vs. miss effects (sSO/SO = sPO/PO > sM/M) or linear changes in brain activity across recognition memory confidence regressors (sSO/SO > sPO/PO > sM/M), which are analogous to a parametric modulations in activity that have been previously associated with familiarity (at encoding: Davachi, 2003; Ranganath et al., 2004; and at retrieval: Daselaar et al., 2006; Montaldi et al., 2006). Finally, graded increases in activity (sSO/SO > sM/M) were observed in the posterior parahippocampal gyrus (domain- general) and left posterior hippocampus (scene-specific) at encoding, left perirhinal cortex (face-specific) during encoding and retrieval, and anterior parahippocampal
gyrus (face-specific) at retrieval only. As these effects showed increases in activity for sure old items compared to missed items, and no difference between sure old and probably old, or probably old and missed items, these effects could not be attributed to either recollection or familiarity.
In the current study there were no significant linear changes in BOLD relating to item familiarity at encoding or retrieval in any of the fROIs. Notably, previous experiments that have revealed parametric fluctuations in familiarity strength have done so across a greater number of confidence response categories (Daselaar et al., 2006; Montaldi et al., 2006; Ranganath et al. 2004; Yonelinas et al., 2005). For example, Ranganath et al. (2004) observed linear changes in brain activity across 5 confidence bins, ranging from high confidence miss to medium confidence hit, within the perirhinal cortex. Likewise, Montaldi et al. (2006) demonstrated similar effects by conducting a parametric analysis across miss and 3 levels of familiarity strength (Fig. 2.12). In the current experiment the number of study trials was restricted (relative to previous experiments) because two stimulus types were employed and fMRI scanning was conducted during the encoding and retrieval phases. Due to low trial numbers, low and medium confidence old responses were concatenated into a single ‘probably old’ regressor. While the aforementioned studies merely accepted linear changes across confidence groups as an index of familiarity, here, due to the low number of response categories, a conservative approach was adopted, whereby sure old had to be significantly different from probably old, and probably old from miss. It is likely that the creation of a single probably old regressor may have masked changes in activity that could have driven a significant parametric modulation using a more liberal approach. This, therefore, could account for the number of instances for which activity for probably old was intermediate between, but did not significantly differ from, sure old or miss.
As discussed in the previous chapter, dual-process accounts of MTL function posit that the hippocampus and perirhinal cortex support recollection and familiarity, respectively (see Section 1.4). Similarly, informational accounts (see Section 1.5), such as binding of item and context (BIC: Diana et al., 2007; see also Davachi, 2006) propose that, while context is processed by the parahippocampal cortex, item-based information is supported by the perirhinal cortex. Item and context information
converges in the hippocampus, where domain-general item-context representations are formed, and reinstated during memory retrieval (Diana et al., 2007; Davachi, 2006). Behavioural estimates of recollection, which requires the recovery of contextual information, should, therefore, correlate with activity in the hippocampus and, according to BIC, in the parahippocampal cortex. Likewise, measures of familiarity, which is supported by item-based information, should be associated with perirhinal cortex activity. Previous functional neuroimaging investigations have revealed that activity in the hippocampus and parahippocampal cortex consistently correlates with correct source, remember and high confidence memory judgements, whereas activity in the anterior MTL and perirhinal cortex is generally associated with memory for single items, know judgements and linear modulations based on item memory (or familiarity) strength (Daselaar et al., 2006; Davachi et al., 2003; Henson et al., 1999; Johnson & Rugg, 2007; Montaldi et al., 2006; Ranganath et al., 2004)
In the current study, although there was activity that pointed to recollection at retrieval in the anterior hippocampus, this was only evident for faces. As noted in the Introduction, however, dual-process and BIC accounts do not make explicit theoretical predictions about stimulus-specificity. As such, it is unclear whether the absence of memory effects for scenes within the anterior hippocampus can be accommodated by dual-process and BIC accounts of MTL function. Furthermore, although there was a significant domain-general encoding effect (posterior parahippocampal gyrus at encoding), the pattern of data could not be ascribed confidently to recollection or familiarity. These data, therefore, provide limited support for dual-process and information-based theories of MTL function.