Building Memories: Remembering and Forgetting of Verbal Experiences as Predicted by Brain Activity
Tóm tắt
Từ khóa
Tài liệu tham khảo
. For reviews see
Halgren E., Smith M. E., Hum. Neurobiol. 6, 129 (1987) ;
; N. J. Cohen and H. Eichenbaum Memory Amnesia and the Hippocampal System (MIT Press Cambridge MA 1993).
Informed consent was obtained from all subjects (seven men five women aged 18 to 29 years). Echo planar and conventional imaging was performed on a 1.5-T GE Signa scanner with an ANMR upgrade. Imaging procedures [see
] included collection of structural images [radio frequency–spoiled GRASS (gradient-recalled acquisition in the steady state) sequence 60-slice sagittal 2.8-mm thickness] and echo planar functional images sensitive to blood-oxygen level–dependent contrast (118 sequential whole-brain acquisitions 16 slices each 3.125-mm in-plane resolution 7-mm thickness skip 1 mm; T 2 *-weighted asymmetric spin-echo sequence: TR = 2 s TE = 50 ms 180° offset = –25 ms).
During each of four scans blocks were ordered: nonsemantic (40 s) fixation (24 s) semantic (40 s) fixation nonsemantic fixation semantic. A brief (8 s) fixation block began each scan. During semantic and nonsemantic blocks 20 words were visually presented: 10 abstract and 10 concrete nouns; half in uppercase and half in lowercase letters. Each word was presented for 1 s followed by 1 s of fixation between words. Subjects responded by left-handed key press. During fixation blocks a cross-hair (“+”) was presented for the entire duration.
Alpha was set to P < 0.05 for all behavioral analyses. Response latencies differed across tasks [ F (1 11) = 89.97]. Memory was assessed in the scanner 20 to 40 m later using a yes-no recognition procedure (8). Subsequent memory differed across tasks [ F (1 11) = 69.50].
Functional runs were averaged within each subject transformed into stereotaxic atlas space and averaged across subjects (8). Activation maps were constructed using a nonparametric Kolmogorov-Smirnov statistic to compare (i) word processing (semantic and nonsemantic) to fixation and (ii) semantic to nonsemantic processing. Peak activations were identified by selecting local statistical activation maxima that were P < 0.001 within clusters of five contiguous significant voxels. These criteria minimize false positives as verified using the logic of control functional runs [E. Zarahn G. K. Aguirre M. D'Esposito Neuroimage 5 179 (1997)].
Subjects were six men and seven women (aged 18 to 35 years). Three additional subjects were excluded because of excessively poor performance. Imaging procedures were similar to experiment one with the exception that imaging was performed using an echo planar T 2 *-weighted gradient echo sequence (3.0-T 128 images TR = 2 s TE = 30 ms flip angle = 90°). During each scan 40 abstract word trials 40 concrete word trials and 40 fixation trials were rapidly intermixed with each trial lasting 2 s. For fixation trials the fixation point remained on the screen for the entire 2 s. For word trials the word was presented for 750 ms followed by 1250 ms of fixation. Abstract concrete and fixation trials were pseudo-randomly intermixed with counterbalancing (each trial type followed every other trial type equally often).
Approximately 20 min later subjects were administered a memory test consisting of 480 studied and 480 unstudied words. Words were presented individually with self-paced timing. Subjects responded “high confidence studied ” “low confidence studied ” or “new.”
An Item Type × Response interaction [ F (1 12) = 22.97] revealed that studied items were endorsed as “high confidence studied” more frequently than were unstudied items [52% and 7% respectively; F (1 12) = 57.04] whereas studied and unstudied items were similarly endorsed as “low confidence studied” [24% and 20%; ( F < 1.0)]. The low confidence response class likely reflects subject guessing and does not differentiate between encountered and novel stimuli.
Encoding task performance was analyzed based on whether the words were subsequently remembered with high confidence (“high confidence hits”) low confidence (“low confidence hits”) or were forgotten (“misses”). Accuracy during encoding was comparable for high confidence hits (88% correct) low confidence hits (88% correct) and misses (89% correct) ( F < 1.0). Semantic decision RTs declined across trial types [ F (2 24) = 9.26]: RTs were longer for high confidence hits (1000 ms) compared to low confidence hits (966 ms) [ F (1 12) = 5.40] which were in turn longer compared to misses (936 ms) [ F (1 12) = 3.91 P < 0.06].
The procedures for selective averaging and statistical map generation for rapidly intermixed trials are described elsewhere (7). Statistical activation maps were constructed based on the differences between trial types using a t -statistic (7). Fixation trial events were subtracted from the word trial events (collapsing across subsequent memory). Miss trial events and high confidence hit events were subtracted from each other as were those for miss trial events and low confidence hit events. Clusters of five or more voxels exceeding a statistical threshold of P < 0.001 were considered significant foci of activation (7).
In addition modest but significantly greater activation for high confidence hits relative to misses was noted in a more ventral extent of left inferior frontal gyrus [Brodmann's area (BA) 47: −34 31 −3] left precentral gyrus (BA 6: −31 0 56) medial superior frontal gyrus (BA 8: −3 28 43) and left superior occipital gyrus (BA 19: −31 −77 34). The superior occipital and medial superior frontal activations can be seen in Fig. 2. Two regions demonstrated less activation for high confidence hits relative to misses: precuneus (BA 31: 3 −43 40) and left middle frontal gyrus (BA 9: −12 31 34). No regions demonstrated greater activation for low confidence hits relative to misses which is in accord with the behavioral data indicating that these two trial types likely did not mnemonically differ.
M. D'Esposito et al. Neuroimage 6 113 (1997). A similar interpretation may be applicable to the results from the blocked-design experiment. However greater left prefrontal activation during semantic processing has been noted even when the nonsemantic processing task has a longer duty cycle [
RTs were matched as follows. First the median RT across all trial types was determined for each subject. Trials with response latencies that fell below the median RT were selected and sorted based on subsequent memory. Selection of trials in this manner resulted in matched RTs for the high confidence hit (852 ms) and miss (839 ms) trial types [ F (1 12) = 2.32 P > 0.15)].
Wagner A. D., Desmond J. E., Demb J. B., Glover G. H., Gabrieli J. D. E., J. Cogn. Neurosci. 9, 714 (1997).
Supported by grants from the National Institute on Aging (AG08441 and AG05778) the National Institute on Deafness and Other Communication Disorders (DC03245-02) the Human Frontiers Science Program and the Deutsche Forschungsgemeinschaft (SFB426). We thank Y. and W. Jacobson and two anonymous referees for helpful comments on an earlier version of this manuscript and C. Brenner and C. Racine for assistance with data collection.