Verbal Episodic Memory Recognition Paradigm

3.9.4 Verbal Episodic Memory Recognition Paradigm

We employed a recognition task to accompany the encoding task in order to measure behavioural performance during recognition and associated functional brain

activation. The task was adapted from an existing task by reducing the number of stimuli in accordance with the shortened encoding task.

3.9.4.1 Theoretical background

We have seen in the introductory section that episodic memory can be tested by tasks measuring free recall or cued recall. Recognition is a type of cued recall that is supported by presentation of probe stimuli that are either targets or distractors. During a yes/not recognition task, a single probe is presented and the subject has to decide if it is a target or not. Recognition tests appear more sensitive at distinguishing AMCI from normal ageing (Anderson et al., 2008; Bennett et al., 2006; Westerberg et al., 2006) and may also be more sensitive than free recall in distinguishing normal ageing from AD because it is less affected by age-related changes (Craik and McDowd, 1987; Parker et al., 2004). These findings suggest that a recognition task would be preferred to a free recall task for our purpose of examining recognition in AMCI, which falls between ageing and AD. Furthermore, both item and associative memory

appears affected in AMCI (Dudas et al., 2005). Taken together these findings suggest that an associative learning task with a yes/no recognition phase may be a particularly sensitive measure to discriminate AMCI from normal ageing.

Recognition can be further divided into familiarity which refers to a feeling of prior exposure to an item without recall of associated contextual information, and recollection referring to the retrieval of the item bound to contextual features such as the time, place or source of experience (Yonelinas, 2002).

The recognition task we used was yes/no implicit associative item recognition task, which followed an intentional encoding phase. The associations are implicit as all the words in a list are semantically related but subjects are not explicitly made aware of this, which enables examination of distinct memory processes such as automatic semantic elaboration.

In summary, these findings suggest the potential of specific recognition tasks to examine distinct memory processes that appear differentially affected in ageing and AMCI. Combining such tasks with functional neuroimaging has the potential to reveal the functional neuroanatomy of recognition processing in controls and AMCI.

3.9.4.2 Experimental design

Recognition was tested after a delay of 12 minutes during which an unrelated attention task and a structural scan were performed. This delay was included to curb rehearsal of studied items because recognition following rehearsal would reflect working memory capacity rather than episodic encoding success. Probe words were presented individually on a screen and subjects used a button to indicate their choice from three possible responses: “Remember” (familiarity with both recognition of the probe and recollection of study episode), “Know” (familiarity in the absence of

recollection of the study episode) or “New” (unfamiliar). The position of choices presented on the screen was similar to the position of the buttons on the button box so that “Know appeared to the bottom right, “Remember” to the top middle, and “New”

to the bottom right. The probe list consisted of 17 target words, 16 novel words and 14 lures. The first and last lists, as well as the first and last words in each of the remaining lists, were excluded from the recognition task to control for the effects of recency and primacy. Targets were typically words in position 2, 5 and 7 in the remaining lists. The lures consisted of the non-presented critical word highly semantically associated with words in each list plus one more non-presented word from the list associated with the lure (Figure 11). Novel words were taken from non-presented word lists. “Remember and “Know” responses were added together as familiarity responses during behavioural analysis. “Remember" refers to recollection of a studied item which is accompanied by conscious retrieval of aspects of the individual’s experience at the time of study, which may include what they thought about or what preceded it, so in a sense being able to relive the moment of study.

Knowing implies only recollection of the studied item (Tulving, 1985). This distinction was included as there is mounting evidence for distinct neural networks involved in remembering and knowing with PFC activity predominantly involved during remembering (for a review see (Knowlton, 1998). The difference between

“Remember” and “Know” was explained to participants, they practiced the task outside the scanner, and were reminded of instructions immediately prior to scanning.

Figure 11. Recognition task.

The figure illustrates the relationship between the critically non-presented highly semantically associated word “hiking” and words that were studied during encoding (blue), and words used as lures (black).

3.9.4.3 Behavioural analysis

For behavioural analysis we calculated corrected recognition rates (Pr) exactly as described in the section above on divided attention. We also calculated a response bias measure (Br) after Feeman and Snodgrass who found normal response bias for patients with isolated amnestic disorders (Korsakoff’s, MTL pathology) but more liberal response bias for patients with AD (Snodgrass and Corwin, 1988).

Br = false alarm rate / (1- Pr)

Subjects might have high levels of familiarity for both old and new probe items if stimuli were not encoded distinctively or when semantic memory representations have been inappropriately activated by prior presentation of target and distractor items.

Stimuli that were not encoded distinctively at presentation could adequately match stored representations of similarly poorly encoded target items or very familiar items in semantic memory, leading to a general increase in recognition responses to both old and new items. A difference in response bias between AMCI and controls could therefore indicate impaired semantic processing.

We also examined whether the groups were comparable on false recognition of lures by calculating the false recognition rate for lures:

Lure Pr = lure HR - FAR

Recognition responses can be classified into four categories (Figure 12). Targets can be correctly identified as previously studied or old (hits) or fail to be recognised (misses). Whereas novel or distractor items can be incorrectly recognised as old (false alarm) or correctly rejected as not studied previously (correct rejection). The two-high-threshold model of recognition memory can be used for interpreting the cognitive processes underlying these four behavioural responses (Snodgrass and Corwin, 1988). This model assumes that recognition involves selection between discrete memory states rather than a continuum of memory strengths as signal

detection models of memory assumes (Snodgrass and Corwin, 1988). Recognition of an item is therefore based on the memory signal trace meeting a specific threshold level. This proposes the existence of distinct memory thresholds for the recognition of old items as old and for the recognition of new items as novel. These two thresholds separate three possible memory states that can be experienced by individuals:

recognition of items as old/familiar, recognition of item as novel/unfamiliar and an intermediate state of uncertainty. Items are correctly recognised as old or familiar when the strength of the memory trace crosses the “high” threshold level. Conversely, if an item fails to activate a memory trace over the “low” threshold it is not recognised

and therefore deemed as new. If the level of memory trace associated with an item is too strong to be novel or too weak to be familiar and therefore ambiguous, then it falls into the uncertain state. Items in the uncertain state are classified as old or new based on the individual’s particular response biases in that situation (e.g. “I must have seen that before” or “If I had not seen that I would remember”). The HR represents the number of times that an individual correctly recognised a target item plus the correct guesses from the uncertain recognition state. The FAR originates from the uncertain recognition state and occurs when a novel item is not recognised as new and a target not recognised as such. Both the Pr and Br are obtained from the HR and FAR which indicates that these measures take into account all the behavioural responses.

Source of Stimuli Figure 12. Recognition response classifications.

Targets can be correctly identified as previously studied or old (hits) or fail to be recognised (misses). Novel or distractor items can be incorrectly recognised as old (false alarm) or correctly rejected as not studied previously (correct rejection).

On a recognition task with equal numbers of targets and novel distractor items, a Br value of 0.5 represents the neutral or chance value indicating that the individual is equally likely to say that an unrecognised item is old or new. A value < 0.5 represents a conservative bias or a tendency to say that an item is new and a value > 0.5

represents a liberal bias or a tendency to say that unrecognised items are old when uncertain.

The functional analysis of this task was event related and required that events of interest be defined and that the timing of each event of interest was required for each experimental run.

Based on the two-high-threshold model of recognition we set up the functional analyses of the recognition data to examine activation during five events:

1. Remember and Know (Old) responses to targets (Hits) 2. New responses to targets (Misses)

3. Remember and Know (Old) responses to lures (False alarm on a lure) 4. New responses to lures (Correct rejection of a lure)

5. New responses to novel and lure distractors (Correct rejection of any distractor)

Examining these conditions in conjunction with the semantic processing manipulation could enable us to draw conclusions about the nature of recognition failure in AMCI.

Table 7 shows the recognition task response data from a sample file for the task at baseline. Probes were presented in the order in which they appear in the table. The timing of presentations was used to create the newstarts file mentioned in the fMRI methods section and identifies the conditions of interest during the analysis process.

Stimuli Word Response Time Stimuli Word Response Time 2 bumble know=correct 364.002 2 cotton know=correct 484.002 1 wing new=correct 372.002 1 needle know=wrong 500.002

Table 7. Recognition task response data at baseline.

The table shows the type of recognition probe (1=lure; 2=target; 3=novel), probes, response, accuracy, and time of presentation from the start of the task for the recognition task at baseline. The probe list consisted of 17 target words, 16 novel words and 14 lures.