Top Banner

Click here to load reader

Perceptual learning of contrast discrimination and its neural · PDF file 2014. 10. 16. · 3.3.3 Psychometric thresholds during the flanker task 156 3.3.4 Reaction times 156 3.3.5

Jul 30, 2021

ReportDownload

Documents

others

and its neural correlates in macaque V4 & V1
Xing Chen
Thesis submitted to Newcastle University
for the degree of Doctor of Philosophy
August 2013
Abbreviations ................................................................................................................. viii
Chapter 2: Roving task ............................................................................................... 117
Chapter 3: Flanker task ............................................................................................... 146
Chapter 4: Control tasks/ analyses .............................................................................. 190
Final discussion and further work ................................................................................. 205
Appendix A: Artifact removal from neuronal data ................................................. 213
Appendix B: Cross correlations between PSTH waveforms of channels............... 227
Appendix C: Characterisation of neuronal tuning properties ................................. 237
Acknowledgements ....................................................................................................... 240
References……………………………………………………………………………………241
Abbreviations ................................................................................................................. viii
1.1.2 Contrast discrimination in human psychophysics studies 2
1.1.3 Electrophysiological signatures of perceptual learning 3
1.1.4 Models of perceptual learning 5
1.1.5 Effects of attention on contrast response functions of visually-responsive neurons 12
1.1.6 Goals of the contrast discrimination task 13
1.2 Neuronal recording methods 15
1.2.1 Data collection 15
1.3 Psychophysics methods 20
1.3.2 Contrast discrimination task paradigm 20
1.3.3 Stages of training on the main contrast discrimination task 21
1.3.4 Measures of perceptual learning 23
1.3.5 Contrast thresholds 25
1.3.6 Reaction times 27
1.4 Behavioural results 28
1.4.1 Perceptual learning with stimuli at the V4 and V1 locations 28
1.4.2 Control task with horizontally-oriented Gabor stimuli at the V4 location 35
1.4.3 Control task with sinusoidal grating stimuli at the V4 location 36
1.4.4 Control task with stimuli of different spatial frequencies at the V1 location 37
1.4.5 Control task with only the test stimulus- not the sample- at the V1 location 37
1.4.6 Discussion of behavioural results from the CD task 38
1.5 Neuronal methods 41
1.5.1 Data processing 41
1.6.2 Non-monotonic contrast tuning functions in V4 61
1.6.3 AUROC/PROBMAT individual channel results 63
1.6.4 AUROC/PROBMAT population results 71
1.6.5 Exclusion of channels with stimulus-evoked suppression 81
1.6.6 Effects of data normalisation 82
1.6.7 Within-trial single-channel correlations in spiking activity 83
1.6.8 PROBMAT and noise correlations 88
1.6.9 Neurometric versus psychometric thresholds 89
1.6.10 Effects of adaptation on stimulus-evoked activity 91
1.6.11 Response adaptation prior to stimulus onset 96
1.6.12 Test-test discriminability 98
1.6.14 Choice probability 104
1.6.16 Discussion of neuronal results from the CD task 109
Chapter 2: Roving task ............................................................................................... 117
2.1 Roving task literature review 117
2.1.1 Stimulus roving during contrast discrimination tasks 117
2.1.2 Insights from a roving paradigm during a bisection task 119
2.1.3 Goals of the roving task 120
2.2 Psychophysics methods 122
2.2.1 Task paradigm 122
2.2.2 Behavioural performance 124
2.3 Neuronal methods 124
2.4 Roving task behavioural results 125
2.4.1 First set of training sessions on a roving task 125
2.4.2 A comparison of performance between non-roving and roving tasks, to monitor task learning
125
2.4.4 Relative changes in performance based on sample contrast 131
2.4.5 Psychometric thresholds during the roving task 133
2.4.6 Reaction times 133
2.4.7 Discussion of behavioural changes during the roving task 134
2.5 Roving task neuronal results 136
2.5.1 Changes in the CRF during training on the roving task 136
2.5.2 Changes in PROBMAT during training with roving stimuli 139
2.5.3 Neurometric thresholds during the roving task 141
2.5.4 Variability of the visual response during the roving task 143
2.5.5 Discussion of neuronal results from the roving task 144
Chapter 3: Flanker task ............................................................................................... 146
3.1 Flanker task literature review 146
3.1.1 Goals of the flanker task 149
3.2 Methods 150
3.2.2 Measures of perceptual learning 150
3.3 Flanker task behavioural results 150
3.3.1 Training on a roving task with flankers at the V1 location 150
3.3.2 Effects of adding flanker stimuli 155
3.3.3 Psychometric thresholds during the flanker task 156
3.3.4 Reaction times 156
3.3.5 Discussion of behavioural results from the flanker task 157
3.4 Flanker task neuronal results 159
3.4.1 Changes in the CRF during training on the flanker task 159
3.4.2 Changes in PROBMAT during training with flanker stimuli 163
3.4.3 Neurometric thresholds during the flanker task 166
3.4.4 Variability of the visual response during training with flankers 167
3.4.5 Discussion of neuronal results from the flanker task 168
3.5 Removal of flanker stimuli 172
3.5.1 Behavioural results 172
3.5.3 Neuronal results 178
Chapter 4: Control tasks/ analyses .............................................................................. 190
v
4.1 Roving task training with matching locations between the two monkeys 190
4.1.1 Methods 190
4.1.2 Results 191
4.1.3 Summary of results from the roving task at the control location 198
4.1.4 Possible differences in task strategy 199
4.2 Spatial attention control task 202
4.2.1 Methods 202
4.2.2 Results 203
Appendix A: Artifact removal from neuronal data ................................................. 213
A.1 Generation of continuously-sampled channel data 213
A.2 Threshold selection for spike extraction using CSC Spike Extractor 214
A.3 Artifact removal 214
A.3.1 Examination of rasters across all recording sessions, for each channel 215
A.3.2 Artifacts induced by the monitor refresh 215
A.3.3 Automated threshold setting to obtain uniform spontaneous activity levels across sessions 218
A.3.4 Artifacts induced by subjects’ movements 221
A.3.5 Inclusion of channels based on the signal-to-noise ratio of spiking activity 225
Appendix B: Cross correlations between PSTH waveforms of channels............... 227
B.1 Methods 228
B.2 Results 231
B.2.1 Cross correlations between PSTHs of channels based on non-roving data 233
B.2.2 Cross correlations between PSTHs of channels based on roving data 235
Appendix C: Characterisation of neuronal tuning properties ................................. 237
C.1 Methods 237
C.2 Results 238
vi
vii
Abstract
We make frequent evaluations of subtle contrast differences in our visual
environment, and often under challenging illumination conditions, whether photopic,
scotopic or mesopic. Our contrast discrimination abilities are rigorously honed from an
early age, and we continue to carry out these fine perceptual judgments throughout our
lifetimes. Thus, the issue of whether substantial improvement in contrast discrimination
is possible during later periods in life, such as during adulthood- and the circumstances
that allow this- has sometimes come under discussion.
Our adult macaque subjects underwent extensive training on a contrast
discrimination task, in which stimuli were positioned at a variety of peripheral and
parafoveal locations. We present clear evidence of contrast perceptual learning at the
behavioural level and show that these changes have neuronal correlates primarily in V4,
rather than in V1. Learning was specific to stimulus location and spatial frequency, but
was transferable across orientations; it took place to a limited degree under stimulus
roving conditions, and could be either facilitated or impeded by the addition of flanker
stimuli, depending on the subject. Upon removal of flankers, levels of psychometric and
neurometric performance returned to their pre-flanker state.
In V4, learning-induced changes encompassed a shift in the point of neurometric
equality and the semi-saturation constant (C50) towards the trained contrast; a decrease
in noise correlations across channels; and an increase in choice probability. In V1,
enhancements in performance were characterised by an increase in spike
discriminability; a shift in the point of neurometric equality and the C50 towards the
trained contrast(s); and a widening in the range and a steepening of the contrast
response function, during the early phase of training. Deteriorations in performance
were accompanied by the reverse effects on V1 activity; furthermore, a general decrease
in V1 firing rates occurred when training was carried out over an extended period of
time, after performance had reached its peak.
viii
AUROC area under the ROC curve
BOLD blood-oxygen-level-dependent
ECG electrocardiogram
IPS intraparietal sulcus
ISI inter-stimulus interval
ix
PSTH peristimulus time histogram
RF receptive field
SF spatial frequency
SNR signal-to-noise ratio
SUA single-unit activity
x
List of tables Table 1. Stimulus parameters used at each stage of contrast discrimination training. ................................ 22
Table 2. Changes in psychometric threshold over the course of training were assessed using a Spearman’s
rank correlation analysis (FDR correction for α-levels: α = .05 × 6/8 = .0375). ............................... 33
Table 3. Differences in performance within individual sessions. For both subjects, when performance was
compared between the first and last 30% of trials, the proportion of correct responses was
significantly higher towards the later part of each session, for stimuli at the V1 location. ............... 34
Table 4. Comparison of subjects’ performance during control sessions, against that seen at the end of
Stage 1. Xmin – Xmax: Ranges of performance seen during late Stage 1 sessions, in which vertically-
oriented Gabor stimuli were presented. Xh: Performance recorded during the single session in which
horizontally-oriented Gabor stimuli were presented. Xg: Performance recorded during the last of the
Stage 3 sessions, in which vertically-oriented grating stimuli were presented. Stimuli were located
at the V4 location during each of these sessions. ............................................................................... 36
Table 5. Number of channels with significant changes for different parameters of the contrast response
function, during training with sample stimuli (monkey 1, V4: N = 29; V1: N = 23; monkey 2, V4: N
= 20; V1: N = 25). An FDR correction was carried out for multiple parameter comparisons. ......... 55
Table 6. Changes in the contrast response function for population activity, with training. A Spearman’s
rank correlation was performed to assess changes in the slope at 30%, the C50, and the minimum
and maximum values, of the CRF. Significant improvements were seen in the slope and the C50 for
monkey 2 at the V4 location, while deteriorations occurred for monkey 2 at the V1 location. A
decrease in the minimum was seen for monkey 1 at the V1 location (FDR correction, slope: α =
.05/4×3 = .0375; C50: α = .05/4×3 = .0375; minimum: α = .05/4×2 = .025; maximum: α = .05/4×1 =
.0125). .................................................................................................................................................. 60
Table 7. Number of channels with significant changes in each parameter of the neurometric function,
during training on the contrast discrimination task (monkey 1, V4: N = 15; V1: N = 21; monkey 2,
V4: N = 11; V1: N = 25). An FDR correction was carried out for multiple parameter comparisons.
.............................................................................................................................................................. 68
Table 8. Results from a paired t-test which compared two different methods of calculating population
PROBMAT values. In both monkeys and at both locations, Pcumulative values yielded better
results than Pmean values, indicating that the pooling of activity across a population of neurons
allowed higher-fidelity encoding of stimulus properties, than merely taking the mean of the
individually fitted parameter values across single channels. An FDR correction was carried out for
xi
multiple comparisons (slope: α = .05/4×4 = .05; PNE: α = .05/4×4 = .05; minimum: α = .05/4×4 =
.05; maximum: α = .05/4×4 = .0125). ................................................................................................. 73
Table 9. Results from a paired t-test, comparing values of each of the parameters derived from AUROC
and PROBMAT methods. The PROBMAT approach yielded higher values for the slope of the
curve at 30% contrast, for both monkeys and both recording locations (slope: α = .05/4×4 = .05;
PNE: α = .05/4 = .0125; minimum: α = .05/4×3 = .0375; maximum: α = .05/4 = .0125; an FDR
correction was carried out as described in the section, ‘Corrections for multiple comparisons,’ on
page 27). The minimum values produced by the trial-wise method were also significantly lower for
both subjects at the V4 location, and for monkey 1 at the V1 location. ............................................ 79
Table 10. Changes in population neurometric functions with training. The PNE for each population of V4
neurons shifted significantly towards the left in both subjects, towards the value of 30%. A
significant increase in slope, as well as a decrease in the minimum value, was also observed for
recordings at the V4 location in monkey 2 (Spearman’s rank correlation, FDR correction, α =
.05/16×4 = .0125). ............................................................................................................................... 80
Table 11. A comparison of population results, before (M1) and after (M2) normalisation of data to the
maximum responses of individual channels. The slope of the neurometric function decreased, and
the minimum value increased after normalisation, for V4 responses in monkey 2 and for V1
responses in monkey 1, indicating that normalisation made the ‘readout’ of population data slightly
worse. Effects of normalisation on the PNE were not consistent across different recording sites. .. 83
Table 12. Spearman’s rank correlation coefficients and q-values, from an examination of changes in
neurometric and psychometric thresholds over the course of training with non-roving stimuli. FDR
correction, α = .05/8×2 = .0125. .......................................................................................................... 90
Table 13. Number of channels where significant differences between test- and sample-induced activity
occurred, when test and sample contrasts differed only slightly. For monkey 1, response adaptation
was seen in around half of the V4 channels (N = 29) and in hardly any of the V1 channels (N = 23),
whereas for monkey 2, adaptation occurred in the vast majority of V4 (N = 20) and V1 (N = 25)
channels. .............................................................................................................................................. 93
Table 14. A Spearman’s rank correlation analysis was calculated to assess whether the differences in
firing rate to sample and test stimuli changed with time. For monkey 1, when stimuli were
presented at the V4 location, adaptation effects decreased with training for the sample contrast
conditions of 31 and 32%, whereas they increased for monkey 2, when stimuli were presented at the
V1 location (FDR correction, α = .05/8×4 = .025). ............................................................................ 95
xii
Table 15. A Spearman’s correlation was carried out to test for changes in population test-evoked spiking
discriminability over the training period, between contrast levels that flanked the value of 30%
(29% versus 31% in V4; 28% versus 32% in V1). ........................................................................... 101
Table 16. Results from a two-factor ANOVA, comparing trial-wise spike variability between early and
late sessions. The Fano factor was found to differ significantly between the two training periods, for
both subjects in both locations (FDR correction for multiple comparisons, α = .05/4×4 = .05). .... 103
Table 17. List of channels for which levels of spiking activity in response to stimuli presented during a
passive viewing task underwent significant changes over the training period. ............................... 108
Table 18. Slopes of the best-fit line to the roving data, shown in Figure 45, for each response conflict
condition. The absolute value of the slope provided a measure of the degree to which performance
changed over the course of training on the roving task. ................................................................... 128
Table 19. Comparisons of performance levels between early and late sessions during training with roving
stimuli, using an unpaired t-test (FDR correction for α-levels, proportion correct: α =.05 × 4/4 =
.05; slope: α =.05 × 4/4 = .05; PSE: α =.05 × 1/4 = .0125; RTcorrect: α =.05 × 3/4 = .0375; RTerror: α =
.05 × 4/4 = .05). ................................................................................................................................. 130
Table 20. Changes in psychometric thresholds during the roving task. FDR correction for multiple
comparisons, α = .05/12×4 = .0167. .................................................................................................. 133
Table 21. Pearson’s correlation coefficients and q-values for correlations between mean RT and session
number. FDR correction, α = .05/12×7 = .0292. .............................................................................. 134
Table 22. Number of channels with significant changes in each parameter of the contrast response
function, during training with roving sample stimuli (monkey 1: N = 23; monkey 2: N = 25). ..... 136
Table 23. Descriptive statistics for a Spearman’s rank correlation analysis to identify changes in the slope,
C50, and minimum and maximum values of the CRF, during training with roving stimuli.
Significant decreases in slope and the maxima occurred for monkey 1, for the 30% and 40% sample
contrast conditions (FDR correction, α = .05/24×6 = .0125). .......................................................... 138
Table 24. Number of channels with significant changes in each parameter of the PROBMAT-versus-
contrast function, during training with roving sample stimuli (monkey 1: N = 23; monkey 2: N = 25,
FDR correction for multiple comparisons). ...................................................................................... 139
Table 25. Statistics for a Spearman’s rank correlation analysis to identify changes in the slope, PNE, and
minimum and maximum values of the neurometric function, during training on roving stimuli.
Significant decreases in slope and increases in the minimum value were seen in monkey 1 for all
three sample contrast conditions (FDR correction, α = .05/24×6 = .0125). .................................... 141
xiii
Table 26. Spearman’s rank correlation coefficients and q-values, indicating changes in threshold over the
course of training with roving stimuli. FDR correction for multiple comparisons for flankerless
data: α = .05/12×4 = .0167. ............................................................................................................... 143
Table 27. Results from two-factor ANOVA, comparing trial-wise spike variability between early and late
roving sessions. Significant changes in the Fano factor occurred over the course of training, in 5/6
cases (FDR correction for multiple comparisons, α = .05/6×5 = .0417). ........................................ 144
Table 28. Comparisons of performance between early and late sessions in the presence of flankers, using
an unpaired t-test. (Student’s t-test, FDR correction for α-levels, proportion correct: α =.05 × 4/4 =
.05; slope: α =.05 × 4/4 = .05; PSE: α =.05 × 1/4 = .0125; RTcorrect: α =.05 × 3/4 = .0375; RTerror: α =
.05 × 4/4 = .05). ................................................................................................................................. 153
Table 29. Changes in psychometric thresholds during the roving task. FDR correction for multiple
comparisons, α = .05/12×9 = .0375. ................................................................................................. 156
Table 30. Pearson’s correlation coefficients and q-values for correlations between mean RT and session
number during training on the roving task with flankers (FDR correction, α = .05/12 = .0042). ... 157
Table 31. Number of channels with significant changes in each parameter of the contrast response
function, during training with flanker stimuli (monkey 1: N = 23; monkey 2: N = 25, FDR
correction for multiple parameters). .................................................................................................. 159
Table 32. A Spearman’s rank correlation analysis was carried out to identify changes in the slope, C50,
and minimum and maximum values of the CRF, during training with flanker stimuli. No significant
changes were seen for individual sample contrast conditions, though a decrease in the minima was
seen for monkey two when data were pooled across conditions (see text for details, FDR correction:
α = .05/24 = .00208). ......................................................................................................................... 161
Table 33. A comparison of CRF parameters between the last third of pre-flanker training and the first
third of flanker training revealed that the addition of flankers had brought about a significant
change across numerous parameters in both monkeys (FDR correction: α = .05/8×7 = .0438). .... 162
Table 34. Number of channels with significant changes in each parameter of the PROBMAT-versus-
contrast function, during training with roving sample stimuli (monkey 1: N = 23; monkey 2: N = 25,
FDR correction for multiple parameters). ......................................................................................... 163
Table 35. A Spearman’s rank correlation analysis was performed to identify changes in the slope, PNE,
and minimum and maximum values of the neurometric function, during training on the roving task
with flanker stimuli. No significant changes were seen for either monkey (FDR correction, α =
.05/24×6 = .0125). ............................................................................................................................. 165
xiv
Table 36. No changes in neurometric thresholds were observed during the flanker task (Spearman’s rank
correlation). FDR correction for multiple comparisons, α = .05/12 = .0167. .................................. 166
Table 37. Results from a two-factor ANOVA, comparing trial-wise spike variability between early and
late flanker sessions. Significant changes in the Fano factor occurred in all cases when flankers
were present (FDR correction for multiple comparisons, α = .05/6×6 = .05). ................................. 168
Table 38. Comparison of subjects’ performance in the absence of flankers, during post-flanker sessions,
and during the end of pre-flanker sessions. Xmin – Xmax: Ranges of performance seen during late pre-
flanker sessions, which took place before flankers were introduced. Xa: Performance recorded
during the last session of post-flanker training, in which roving stimuli were presented, after the
removal of flankers. ........................................................................................................................... 176
observed throughout non-roving and roving training (FDR correction for multiple comparisons, α =
.05/12×6 = .025). ............................................................................................................................... 186
observed throughout non-roving and roving training when stimuli were positioned at the V1
location, though this was true for more parameters in monkey 2 than in monkey 1 (FDR correction,
α = .05/12×7 = .0292). ....................................................................................................................... 187
Table 41. Positive correlations between z-scored neurometric and psychometric function parameters were
present throughout non-roving training for monkey 2, though not for monkey 1, when stimuli were
positioned at the V4 location…