側膝核在特殊閱讀障礙兒童與成人所扮演的角色

(1)

行政院國家科學委員會專題研究計畫成果報告

側膝核在特殊閱讀障礙兒童與成人所扮演的角色

研究成果報告(精簡版)

計畫類別：個別型

計畫編號： NSC 99-2410-H-006-044-

執行期間： 99 年 08 月 01 日至 100 年 07 月 31 日執行單位：國立成功大學認知科學研究所

計畫主持人：龔俊嘉

共同主持人：趙垂勳、陳欣進、曾世杰、沈戊忠計畫參與人員：學士級-專任助理人員：陳秋月

碩士班研究生-兼任助理人員：葉丁瑞碩士班研究生-兼任助理人員：Mahen Nade

報告附件：出席國際會議研究心得報告及發表論文

公開資訊：本計畫可公開查詢

中華民國 100 年 10 月 31 日

(2)

中文摘要：利用功能性磁振造影(fMRI)，我們欲探討一個在閱讀障礙裡一個尚未深入探討的可能成因─視覺系統缺損─尤其是在視覺傳導路徑的第一個中繼站：側膝核中的巨細胞層(magnocellular layers)─所導致。此種閱障學童約佔閱讀困難人口中的 10%，

屬於較為少見的族群。文獻中對此種巨細胞缺損所導致閱讀障礙的真實性仍處於爭議階段：非但由於對此種功能性缺陷的支持大多來自間接的心理物理或死後解剖，缺少直接的證據；也由於此種假說所解釋的閱障比例較小，生態解釋力弱(相較於目前主流的語音覺識與神經生物假說)，因此是一個相當具挑戰的題目。欲便利資料蒐集的速度，將與中南部閱讀障礙的專家合作，藉由他們篩選出的閱障兒童及成人，進行多點(包括台中，

台南，高雄，及台東等)的行為施測(包括基本的閱讀能力測驗與至少兩種視覺作業:點協調與對比敏感阈值作業)，再將點協調作業中反應阈值高於非視覺障礙閱障(N=20)與正常控制組 (N=20)兩標準差以上的閱讀困難兒童與成人(N=20)，轉介至高雄與台中進行 3T fMRI 實驗。預計每位受試者會做三次 fMRI 實驗，共同的目標區是在利用高解析度(1.5x1.5x2 mm3)的側膝核影像，探討在高低對比(high and low contrast)，不同空間頻率 (spatial frequency)，與不同時間頻率(temporal frequency)下，巨細胞(magnocellular)與微細胞層(parvocellular layers)的血流變化反應(組間與年齡間)比較。

英文摘要： Developmental dyslexia is one of the most challenging problems facing psychologists and educators, with about 5%-15% estimated incidence rate worldwide. Among its top three explanations, including phonological awareness, neurobiological approach, and the magnocellular LGN deficits, the last hypothesis is probably the most controversial. Despite of the flourishing developments in neuroimaging methodology, the direct evidence supporting or refuting the magnocellular hypothesis is still lacking. In the present proposal, we seek to use high-resolution fMRI to look into this tiny structure, testing the derived expectations from the magno-deficit hypothesis. We will first implement the behavioral tests, including reading, working memory, and several perceptual tasks on 40 referred dyslexic children (half with heightened motion coherence threshold, 20 without) and adults, along with age-matched controls, followed by a series of fMRI experiments covering contrast, spatial frequency, and temporal frequency manipulations. The quantitative measures of magno and parvo-LGN response properties will be assessed for or against the magnocellular hypothesis. The second sets of experiments will manipulate spatial attention, comparing the effect of attended vs. unattended condition on the magno- and parvo-LGN. We hope to provide insights into the complicated processes from the deficits in early processing pathways to the high- level reading comprehension, with the clearer evidence for both the

(3)

pathology of at least subtype of dyslexia, and whether LGN plays a role in affecting these dyslexic subjects.

(4)

NSC Report Outline:

Introduction

Literature Review/Objectives Materials and Methods Results

Conclusions/Future Work Figures

References Introduction

Previous work on the lateral geniculate nucleus (LGN) has shown that it exhibits a number of interesting properties. [1] found that attention modulated responses in the LGN much like it did in the visual cortex: larger signal associated with attended vs.

unattended and increases in baseline signal in anticipation of stimulus onset. [2] found correlations between the LGN signal and subjects’ subjective percepts in a binocular rivalry. An overview of the LGN as more than a gateway to the visual cortex has been provided in [3]. Whilst the low-level bottom-up (or stimulus driven) signals have been studied extensively in the LGN, few studies have attempted to examine top-down influence, leaving a fair bit of knowledge of such effects to be found. Our study attempts to ascertain the relevant input of general top-down effects as found in responses, as opposed to stimuli. Further, we hoped to gauge the relative distribution of these signals in the visual cortex and the LGN, specifically across voxels

comprised of different cell-types.

Literature Review/Objectives

Of particular interest and relevance are the results from high-resolution studies that probed the different cell-layer types in the LGN. The LGN can be divided into two main layers – magnocellular (M-cell) and parvocellular (P-cell). [4] was the first study to properly attempt to delineate M- and P-cells in an fMRI setting. The LGN is the first stop for signals travelling from the retina on to the visual cortex. While previously thought of as a gateway to the visual cortex, it has displayed many properties akin to the visual cortex, including top-down influence. Finding parallels between visual cortex and the LGN is thus important in this field, as it expands our knowledge of the working visual system. One important finding in the visual cortex is that of [5, 6] - BOLD signal in the primary visual cortex can be modulated by

responses to a larger extent than previously assumed. They used a difficult, low- threshold contrast detection task where subjects saw a noise annulus and judged whether they saw a grating pattern or not. As a signal detection paradigm, it is assumed that at low, difficult-to-detect contrasts, the population of target-present (noise pattern and grating) signals would overlap with the population of target-absent (only noise pattern) signals. Thus, data was analysed based on hits, misses, false alarms and correct rejects. The average signal in parts of the visual cortex that corresponded to the annulus was found to be higher in false alarms than in misses, suggesting a large top-down response bias (i.e. the target that was present in miss trials evoked a smaller response than trials where subjects thought they saw a target

(5)

which was not there). Using a novel ROC analysis method (for an example of the ROC method in fMRI, see [7]), we attempted to extend this finding in the visual cortex by looking for selectivity in voxels, and find it in the LGN.

Materials and Methods

Our study required localization of the lateral geniculate nucleus and retinotopic

mapping of the visual cortex prior to the noise paradigm. Further, scans were acquired at different sites (Kaohsiung Medical University and Chinese Medical University Hospital), and were thus acquired using different parameters. A brief description of necessary differences and results can be found in Table 1. Subjects underwent several sessions. Localisation sessions and noise paradigm sessions were separated. An anatomical scan was collected in most sessions; three were collected during each retinotopic session for inflation and flattening. All experiments were run on a MacBook Pro with MATLAB® and the Psychophysics Toolbox ([8, 9]), using in- house scripts and functions. All functional data was analysed with AFNI ([10]).

Anatomical data was registered to slice-time-corrected, motion-corrected functional volumes temporally close to their acquisition time (using a Local-Pearson Correlation cost function, as developed in [11]), unless too much motion was detected, in which case they were registered to the functional volumes with average maximum

displacement. Further analysis was conducted with in-house MATLAB® scripts, which also utilised the “Afni-Matlab” toolbox (http://afni.nimh.nih.gov/afni/matlab/).

Anatomical skull-stripping, segmentation, parcellation, inflation and flattening (after manual cuts) were performed by Freesurfer 5.0 and 5.1's “recon-all” pipeline ([12, 13]). Subsequent viewing and ROI-drawing on surfaces were conducted with AFNI and SUMA [14].

Scan Parameters KMU

32 axial slices, 3.4mm x 3.4mm x 4mm. TR 2s, TE 30ms, flip angle 90 degrees.

Images were projected into goggles.

CMUH

26 oblique slices, 1.5mm x 1.5mm x 2.4mm (originally coronal, later axial partial coverage focussed on inferior brain areas critically including LGN and visual cortex).

TR 2s, TE 35ms, flip angle 90 degrees. Images were projected onto a piece of perspex in the scanner bore by a projector placed in the back of the room.

LGN Localization

Our method followed those established in Schneider et al. (2004). Studies that have localized the LGN rely on stimulating the left and right LGNs using preferential left and right hemifield checkerboards. For the purposes of delineating m- and p-cell candidate voxels, we used 6 high-contrast (100% contrast) and 8 low-contrast (10%

contrast) runs in a single session. Each alternating hemifield lasted for 16s, with 16 alternations, making for 256s a run. Subjects were required to stare at a fixation cross and press a button whenever the cross flashed. Due to inadequate results from high- contrast runs, no low-contrast runs were pursued in KMU (Fig. 1). The following analysis discussion thus applies only to data acquired at CMUH.

In addition to the standard preprocessing mentioned earlier, no extra spatial or magnitude normalisation was done. However, signal was normalised to percentage

(6)

signal change with the last two time-points of the previous trial used as a baseline for result plotting. Pilot results from coronal scans in one subject were promising, but the move to oblique axial slices was made so as to include the visual cortex. Separate linear regressions were run for each subject in original space for high and low- contrast runs respectively, as well as together. However, results for a “left annulus – right annulus” contrast (Fig. 2 & 3), at a corrected p < 0.01 showed abnormally large clusters that extended far more posteriorly than expected by knowledge of anatomical location and previous results from PI's earlier work. As a complement then, the averaged high and low-contrast results were Fourier-transformed, resulting in phase, amplitude, and a correlation coefficient at each voxel, as per [4]. This analysis consistently yielded almost no LGN voxels (Fig. 4) at the expected correlation coefficient (r > 0.25; p-values at this coefficient value differed between subjects, but were usually p < 0.05), even when combined for power. Whilst it is assumed that the high- and low-contrast activate different populations of voxels, a significant overlap is still expected. Thus far, a few tentative ROIs have been drawn (see Fig. 5), where we have attempted to localise the LGN through the localiser results, anatomical

knowledge, and average time-course plotting (Fig. 6). However, the areas cannot be taken as precise. Time-course plots (averaged over time and the ROI voxels) indicate that the displayed ROI is largely correct, but plots of the average time-courses of individual voxels in the ROI were not wholly consistent. Further, the pattern across the entire ROI was only consistent with high-contrast runs, not low-contrast runs. Due to an inability to distinguish LGN and non-LGN voxels, contrast modulation indices (CMI) to delineate M- and P-cells have not proven fruitful.

Retinotopic Mapping of Visual Cortex

Originally (at KMU), we acquired polar angle and eccentricity maps for retinotopic mapping with the intention of using relevant eccentricity measurements to restrict our analyses to the annulus area. Stimuli consisted of an expanding coloured

checkerboard ring, with one complete cycle occurring every 32s. However, since early visual areas can be delineated from just polar angle mapping and we’d designed different annulus localisers, we restricted further sessions to just polar angle mapping.

Originally, a coloured rotating checkerboard wedge completed a 360-degree cycle every 32s, with 16 cycles. To increase the effect of the travelling waves, this was changed to 8 cycles of 64s each. Each subject performed 3 – 5 runs (depending on scan site). Analysis was conducted according to established methods ([15]), which included Fourier transforming the data to acquire phase results. Results were

confirmed by AFNI’s RetinoProc pipeline. Results from both sites were adequate for delineation of primary visual cortex areas V1/V2/V3 upon flattened cortical surfaces (see Fig. 7). Subject CCK’s retinotopic maps were done at neither KMU nor CMUH, instead they were taken from earlier pilot results during PI’s earlier work (Fig. 8).

Annulus Localization

Several attempts have been made to localise the area of visual cortex specific to the noise paradigm annulus, each showing varying and inconsistent levels of success. As per ([5]) a block design was tried, with 12s of an annulus and 12s of fixation. As per previous results by the PI at a different scan site, an alternating noise hemifield (equal in size to half the annulus without the vertical meridian) of 12s per side was also tried.

Lastly, results from the actual noise paradigm were used in a linear regression

(7)

analysis. An earlier version of the first design (with shorter 8s annulus, 8s fixation blocks) was partially successful in one subject, giving partial annulus eccentricity.

Subsequent efforts have, however, failed.

Noise Paradigm

The noise paradigm was a slow-event related design adapted from the work of Ress, Heeger and colleagues (see [5, 6]). Each trial consisted of a 1s noise annulus with a 50% chance of the grating pattern being present, followed by 15s fixation to allow for the haemodynamic response to return to baseline. Subjects typically performed 8 – 10 runs of 20 trials each per session, with the aim of acquiring at least two usable

sessions per subject (for 320 - 400 trials in total). Subjects had extensive behavioural training to determine their contrast threshold, with the requirement that their accuracy (sum of hits and correct rejects divided by total trials) was between 65 and 75%

without sacrificing false alarms. Behavioural thresholds ranged from 1.6 to 2.0%;

thresholds in the scanning session were increased by 0.4 – 0.6% to accommodate viewing conditions and discomfort. Table 1 shows representative behavioural results from scans at both KMU and CMUH. No noise paradigm analysis has been completed for the CMUH data. KMU data was preprocessed as described above, before being convolved with a gamma-variate HRF for linear regression analysis. Not only did this help with quality checks, it also allowed for linear detrending as per the recommended method (modelling of linear trend in regression analysis, instead of detrending before analysis), which allowed us to extract “cleaned” signal. This was then used in a voxel- wise ROC area-under-the-curve (AUC) analysis in the following manner: single sessions were sorted into 4 response categories (Hits, Misses, False Alarms, and Correct Rejections), the third and fourth time points (corresponding to the BOLD signal peaks at 6 – 8s) of each trial in each category were converted to a percentage of the last two time points of the previous trial (baseline), and the resulting arrays were analysed using a custom ROC MATLAB function. Typically, an ROC analysis works out the selectivity of a receiver (here, a voxel) between a target-present (Hits and Misses) and target-absent (False Alarms and Correct Rejections) population [ref]. We modified our analysis to do both this and work out the selectivity between a response- yes (Hits and False Alarms) and response-no (Misses and Correct Rejections)

population, so informing us about a voxel’s selectivity for response-influenced signals. An AUC value of 0.5 denotes chance performance, meaning a voxel does not differentiate between the two tested category populations. Given the difficulty of the task and the sparseness of trials, we looked for AUC values greater than 0.56. Results from the target AUC (AUCt) and response AUC (AUCr) were permuted for

significance testing.

Results

It must be noted that the following results are incomplete. Results from KMU were whole-brain, but due to incomplete knowledge, were not detrended. Not all results were consistent across subjects. Target selectivity (AUCt > 0.56) was found largely in early visual areas across all subjects. Some subjects showed strong response

selectivity (AUCr > 0.56) in posterior parietal and prefrontal areas. Some response- selectivity was found in the visual cortex. While the specific distribution varied across subjects, the results were strong, with clusters of voxels present for each AUC-type

(8)

(Fig. 9 & 10). This in itself is not surprising, and is consistent with results from the referenced studies.

Conclusions, Problems & Future Work

To the extent that we could replicate results from [ref], the visual cortex results were unsurprising. They point to top-down effects in the visual cortex, which we expect to find in the LGN. Because of problems in the localisation of the LGN, none of our LGN hypotheses have been confirmed. The goggle display system in KMU suffered from many problems, most important of which were diminished and blurry screen displays. This explains the generally higher contrast thresholds in the noise paradigm (see Table 1). While the conditions at CMUH, especially display-relevant are much better, there are a few problems still to be worked through. Our decision to use high- resolution fMRI was motivated by a need to delineate M- and P-cell voxels.

Consequently, only partial coverage was possible, and at a signal-to-noise ratio (SNR) loss. Although robust activation in areas including and around the LGN was found, neither regression nor Fourier transforms provided adequate information for proper localisation. However, recent discussions with the operator have focussed our

attention on possible RF interference from our display setup (projector) and response box. Further, subjects have reported scanning interfering with the display by creating lines that travel down the display. These were not visible during the earlier

localisation scans (i.e. they have only appeared recently). Our current solution is to Fourier-transform our raw signal, isolate the RF interference, and reanalyse the data.

Should this fail to produce well-localised results, our final attempt will be to draw large ROIs encompassing the LGN and plot time-courses for all high- and low- contrast conditions of interest. In addition to SNR loss, the signal had been affected by inconsistent homogeneity. Further, our protocols have been hampered by scanner overheating (due to the taxing nature of our high-resolution protocol) and final DICOM volumes often not being written to disk.

Whether or not the LGN localisation succeeds or not (leading to further analysis), final results will include visual area analysis. Due to concerns regarding the size of our samples (low number of trials) and the sensitivity of our ROC AUC

measurements ([16]), we plan to implement a surface-based MVPA ([17]) using the

“Surfing” toolbox (http://surfing.sourceforge.net) for MATLAB for visual cortex analysis. At least one study has shown that surface-based MVPA is more sensitive to local information content and provides better spatial selectivity than volume-based MVPA. Another method of interest that may find a place in our final analysis is AFNI’s ANATICOR ([18]). Typically used for resting state data, it attempts to

remove transient machine artifacts that are not obviously present when eyeballed. Due to our low SNR, any method that removes noise without affecting stimulus-related signal must be considered.

(9)

Figures

Fig 1. – Inadequate LGN localisation

Fig 2. Subject ANG – high- contrast LGN- localiser results, p

< 0.001 (corrected)

Fig 3. Subject MN – high- and low- contrast LGN- localiser results, p

< 0.001 (corrected)

Fig 4. Subject MN – high-contrast Fourier transform result, r = 0.25, p < 0.02 (uncorrected)

(10)

Fig 5. Subject YDR – tentative left and right LGN ROIs

Fig 6. Subject YDR, results from high- contrast runs in right-LGN ROI.

Fig 7. Subject YDR, right hemisphere polar angle maps.

Fig 8. Subject CCK, right hemisphere visual area delineation

(11)

Table 1 (KMU = columns 2 – 5; CMUH = columns 6 – 9)

Name Sessions Accuracy H/M/FA/CR Contrast Sessions Accuracy H/M/FA/CR Contrast

CCK - - - - 2 70% 126/69/38/155 2.1-2.4

MN 2 70% 103/57/17/143 2.9-3.3 2 61% 116/63/67/102 1.8-2.0

YWS 3 61% 168/91/101/159 2.9-3.0 - - - -

CCY 2 65% 82/73/27/127 2.5-2.8 1 64% 52/48/25/75 2.0-2.2 YDR 2 61% 136/39/26/148 2.6-3.3 1 60% 54/36/35/54 2.0-2.2

ANG - - - - 1 50% 66/34/65/35 1.6-2.2

Fig 9. Subject MN, right hemisphere, AUC results. Green clusters are AUCr >

0.56; blue clusters are AUCt > 0.56.

Fig 10. Subject YDR, left hemisphere, AUC results.

(12)

References:

1. O'Connor, D.H., et al., Attention modulates responses in the human lateral geniculate nucleus. Nat Neurosci, 2002. 5(11): p. 1203-9.

2. Wunderlich, K., K.A. Schneider, and S. Kastner, Neural correlates of

binocular rivalry in the human lateral geniculate nucleus. Nat Neurosci, 2005.

8(11): p. 1595-602.

3. Kastner, S., K.A. Schneider, and K. Wunderlich, Beyond a relay nucleus:

neuroimaging views on the human LGN. Prog Brain Res, 2006. 155: p. 125- 43.

4. Schneider, K.A., M.C. Richter, and S. Kastner, Retinotopic organization and functional subdivisions of the human lateral geniculate nucleus: a high- resolution functional magnetic resonance imaging study. J Neurosci, 2004.

24(41): p. 8975-85.

5. Ress, D. and D.J. Heeger, Neuronal correlates of perception in early visual cortex. Nat Neurosci, 2003. 6(4): p. 414-20.

6. Ress, D., B.T. Backus, and D.J. Heeger, Activity in primary visual cortex predicts performance in a visual detection task. Nat Neurosci, 2000. 3(9): p.

940-5.

7. Pessoa, L. and S. Padmala, Quantitative prediction of perceptual decisions during near-threshold fear detection. Proc Natl Acad Sci U S A, 2005.

102(15): p. 5612-7.

8. Brainard, D.H., The Psychophysics Toolbox. Spat Vis, 1997. 10(4): p. 433-6.

9. Pelli, D.G., The VideoToolbox software for visual psychophysics:

transforming numbers into movies. Spat Vis, 1997. 10(4): p. 437-42.

10. Cox, R.W., AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res, 1996. 29(3): p. 162- 73.

11. Saad, Z.S., et al., A new method for improving functional-to-structural MRI alignment using local Pearson correlation. Neuroimage, 2009. 44(3): p. 839- 48.

12. Fischl, B., M.I. Sereno, and A.M. Dale, Cortical surface-based analysis. II:

Inflation, flattening, and a surface-based coordinate system. Neuroimage, 1999. 9(2): p. 195-207.

13. Dale, A.M., B. Fischl, and M.I. Sereno, Cortical surface-based analysis. I.

Segmentation and surface reconstruction. Neuroimage, 1999. 9(2): p. 179-94.

14. Saad, Z.S. and R.C. Reynolds, Suma. Neuroimage, 2011.

15. Sereno, M.I., C.T. McDonald, and J.M. Allman, Analysis of retinotopic maps in extrastriate cortex. Cereb Cortex, 1994. 4(6): p. 601-20.

16. Hanczar, B., et al., Small-sample precision of ROC-related estimates.

Bioinformatics, 2010. 26(6): p. 822-30.

17. Oosterhof, N.N., et al., A comparison of volume-based and surface-based multi-voxel pattern analysis. Neuroimage, 2011. 56(2): p. 593-600.

18. Jo, H.J., et al., Mapping sources of correlation in resting state FMRI, with artifact detection and removal. Neuroimage, 2010. 52(2): p. 571-82.

(13)

2011 APCV 心得

今年的 APCV 2011 in HK, 是第一次帶領學生一同參加。畢竟參加過一兩次國際會議後，有學生一同參加，非但對自己而言能夠消除旅途中的寂莫，看著學生的學習成長也是一種正向的學習經驗，感覺也比自己一個人參加來得有趣。第一天約十點多到場，聽到的是 perceptual learning 的 talk series，

遇到熟識的黃立強老師與其他來自世界各國的老師報告。大部分的是行為實驗，但聽到一些如"attentional awakening"，並強調與"attentional blink"的不同歷程，讓人回想起以前認知心理學家的細膩思維。此外下午也有sound- induced visual illusion （如在一個visual flash裡撥放兩個 beep，會造成兩個 flashes 的錯覺，Shams 2000 Nature），作者利用 neural network 的方式 model 在 LGN 的 magno/parvo neuron 如何能 integrate 不同時序與空間向度的兩種訊息。最後，Jiye Kim of USC 報告了與中大阮啟弘老師的合作，利用 TMS 與 fMRI 看物體interact 時LOC/PPA 等的反應是否比 separate 時來得大(fMRI yes, but no in TMS by blocking V5)，也得到了 student travel award。

第二天上午的 keynote speaker 為 Harvard/MIT 的 Jeremy Wolfe 教授，專長在 visual search，近來在homeland security 的計畫也讓人見識到基礎研究做得好後，會如何能 benefit applied research。第二天下午的視覺與藝術的section算是相當令我印象深刻。不只因為這是最近國科會人文處計畫中的一個讓我們覺得無處下手的子主題（所以很想知道目前的專家看法為何），

另外也如同介紹人 Bianca 所說，大部分研究者都有他們的 day job，所以慢慢累積興趣，一小點一小點的慢慢做的方式，與我目前在神經經濟學與神經心理的興趣歷程很像。第一個 speaker 提到文藝復興早期的深度知覺做法：

如Duccio 的 Maesta，或是Giotto 的 chapel 等；早期如這裡的錯覺做為一個 hand-out，就讓我們可以看非常久。此外，Jackson Pollack 的 Fractal 也做為一個聯結自然視覺與美學的中介。Jim Enns 談到林布蘭所畫左右眼的不同細節程度，似乎代表了人注視的一種自然傾向：較清晰的細節較吸引人們的注視，也較讓我們感到喜歡(attention and liking are linked)。週六晚上的 banquet 還看到著名的川劇變臉，更是讓人大開眼界。

第三天上午 Martin Banks 的 keynote 由於無空調的關係，無法集中精神，

但下午的 symposium on predictive coding in V1 etc 算是相當有趣的topic。

聽到了一兩個非常有趣的 talk，如同操弄 orientation 與 information，分離兩個有趣的問題，個人便猜想其一定是 submit 到 nature（而且被接受的程度很高），主持人Lars Muckli 也報告了數個有趣的 fMRI study（包括發表在 JNeuro & PNAS 的兩篇，利用似動運動的方式來預測經過的區域會有較

(14)

低的表現等等）。第四位講者用的是"The good, the bad, and the ugly" 最末段裡三位槍手彼此的 recursive inference 去推論我們theory of mind 的感知極限。晚上去看 Peak Tram，同時也是曾加慧老師的一個有趣的 poster

work。結果哩，感覺真的有（很多遊客也都感受到），不管是上去下來都有，

而且似乎晚上較為清楚。Poster 的結論是這個illusion 有身體感知

(verstibular)的成分，知覺傾斜(perceived slant)的成分，知覺運動的成分，

所以是一種 multi-sensory的錯覺。據說他們為做這個實驗，坐了兩百多次的 peak tram（一次要六十五港元）。我們都建議在發表後跟香港旅遊局收取廣告費（或受試者費）贊助。學生 Mahen 參加足球比賽，聽說跑得腰酸背痛。

第四天，也是最後一天的行程，從上午的Face & Bodys，不管是 other-race effect (廣州中山大學的高國梅教授），一位從京都大學來的老師講 anger- superiority effect，從演化上避凶優勢解釋，到其在倒立效應上的差異等。

Queensland Univ. 的 Guy Wallis 提到當把 schematic faces 轉成構成的 line drawing 時，一些處理的優勢仍然存在的有趣現象（似乎必須尋找比

facedness 外更好的解釋）。Dartmouth College 的 Meng Ming 的學生 Yang Hua 提到 left- vs. right-FFA 對於 faceness 的 pictures 似乎有不同的反應方式，建議了一個非常重要的，與我們實驗室自己過去的像臉物體的 fMRI data 有著有趣的 implication（也希望學生回去能重新分析這些資料，看是否有類似的結果）。從southern cross univ. 的講者提到了用 point-like walker 的 exp 進行性別(sex vs. gender)的判斷，加上走路聲音頻率的操弄或嗅覺上不同性別的汗液等，也都可以得到有趣的 bias 效果。最後的講者提到了 Bovtinick & Cohen (98)的 nature article on rubber hand illusion，主要是指看到別人用毛筆刷假手時，看到的人自己的對應手也會有相應癢的感覺。他進一步操弄鏡像的遠近與第一或第三人角度的效果。下午的

perceptual learning 是一個相當有意思的section。主講人于聰（北師大）與方方（北大）都做了相當精采的演講：前者總結了目前對PL的傳統觀點，並基於他個人的研究心得，提出了為何先前一些相當generalization 相當有限的PL 研究為何如此：如果在訓練時能讓學習者知道上層(abstract level)的規則，則 generalization是較有可能發生的（後來的 Philip Kellman 更進一步擴展此一想法，將其應用到了數學與醫學教育的層面。後來與李金鈴老師提到如此的感動時，也提到不同年紀的研究者所關切的畢竟仍是不太一樣。

年輕的學者較著重在實徵或方法的研究；年長的會把角度放長遠，更重視其理論與應用層面等等）。第三位的Shawn Green 目前仍在做 Postdoc，但因其數篇Nature 文章的大受重視（如打電玩能影響知覺與手眼協調等），

提出一個從 Bayesian 的角度的抽象學習法則。打電玩的人如果能從其電玩經驗中學到抽象思考的 rule，則一樣可以 apply 到 attention blink，或

(15)

dot coherence task 如此的知覺作業中。總之，下午的section 是相當令我們

（包括 Alan Wong，Janet Hsiao 等）這些在 percetual expertise network 下成長的人感到相當的興奮。下午的 poster session 多是在傳統知覺

domain，還看到了陳建中教授所指導的一位北一女資優班高二生的poster。

台灣的教育真是愈做愈精緻！

嚴格說來，這次在香港參加的視覺年會，雖然不能算是大型的國際會議（並無固定的組織。參與人數也多加200-300 人之間），但主辦單位仍是努力以赴，我們也感受到他們在細節上的用心（如舞龍舞獅，校長致詞，川劇變臉，足球比賽，海釣烏賊，香江夜景等），全部的參與實驗室人員都很喜歡。

香港人一般皆很有禮貌且樂於助人（我們住的地方北角的hostel 雖很便宜，

安全與附近的食物皆不錯）。八達通與電話預付卡，讓我們的行動與聯繫不受限制。下一次的仁川會議，真是令人期待！

From: "APCV 2011 organization committee" <[email protected]>

Date: April 19, 2011 3:34:50 PM GMT+08:00 To: <[email protected]>

Subject: APCV 2011 Abstract Submission Dear Chun-Chia Kung

Thanks for submitting your abstract to APCV 2011, Hong Kong. We are pleased to inform you that your submission (Voxel size, dependent measures, or similarity to faces? Review and comparison of various possibilities on explaining the mixed FFA-expertise correlation results) has been accepted for presentation at the conference as a poster. We hope to circulate the schedule, which will inform you of the date and time of your presentation, within the next two weeks.

Student travel award applications are still being reviewed. If you asked to be considered for such an award, you will be notified of the outcome within the next two weeks.

Please be reminded that the deadline for early online registration is May 15, 2011.

Registration is assumed only after the payment is made. Payment status can be enquired on the online meeting system. To register for the meeting, go to

http://www.apcv.net/register.php .

If you have any questions about your presentation, please contact us at [email protected].

(16)

We look forward to seeing you in Hong Kong in July.

APCV Organizing Committee [email protected]

(17)

Revisit the FFA-expertise correlation debate: the effect of the FFA localizer and the evidence of

middle fusiform gyrus as modulated by task load, stimulus familiarity, and expertise

Chun-Chia Kung, Chiu-Yueh Chen, and Citta Yu-Jen Tsai

Department of Psychology and Institute of Cognitive Science, National Cheng Kung University, Tainan, Taiwan

Background

Methods

Participants: N=17 (10 RI bird experts, 7 novices)

Mean age for subjects was 44 ± 2.5 yrs (mean age for

bird experts: 47.2 ± 16 yrs, bird novices: 28.1 ± 5.14

yrs), and the mean years of birding experience for the

10 bird experts was 11.3 ± 5.6 yrs.

Stimuli Four sHmulus categories were deﬁned: 200

Max-‐Planck InsHtute face database faces, 200 common

objects, 120 Rhode Island and 80 Asian birds.

Behavioral task measuring bird expertise (d’)

fMRI tasks passive viewing, 1-back identity, 1-back

location, and 2-back identity (crossed with 4 stimulus

classes; faces, objects, RI birds, Asian birds.

fMRI procedure

-  1.5 T Siemens Symphony, single head-coil, at

the MHRI

-  Blocked design. 16 s per block, TR = 2s, TE =

38 ms, 25 axial slices, gap = 0, FOV = 192 mm ² ,

and the voxel resolution =3 x 3 x 5 mm ³

-  2 runs for each task, counterbalanced across Ss

-  Task accuracy and reaction speed both

emphasized

Results

-‐  TheoreHcal signiﬁcance of category-‐selecHvity

area, such as “FFA: faces or experHse”, has been

a subject of extensive debate

-‐  Even so many studies doing FFA-‐experHse

correlaHon, whether non-‐face objects of

perceptual experHse within the FFA correlate

with their behavioral experHse is sHll unsolved

-‐  From Table 1 (leU), we idenHfy 3 untested

factors: localizer task, experimental task, and

other-‐species manipulaHon, in the current study.

!

Table 1. Selected published studies on tesHng natural experts’ FFA acHvity and

their behavioral experHse index. The highlighted columns represent the factors

addressed in the current study.

Figure 2. The distribuHons of subject’s bird

experHse index (d’) in both expert and novice

groups. Even though there are some overlap

between the two groups, the mean diﬀerence is

highly signiﬁcant (t(15) = 4.55, p<.001).

Figure 1. Examples of the tasks and sHmuli. This ﬁgure also illustrates the

structure of a single scanning run which contained 4 blocks corresponding

to the 4 sHmulus condiHons (blocks were interleaved with ﬁxaHon).

(a) Behavioral responses: overall, main eﬀect

of sHmuli, task, and interacHon. Faces are

hardest, followed by birds, then objects

(b) Average BOLD responses in rFFA: no

signiﬁcant cross-‐task diﬀerences, and is not

the main focus here

fMRI data analysis

- BrainVoyager QX

•  Preprocessing: slice-time correction, motion

correction, temporal smoothing (< 3 cycles/s)

•  GLM for Faces (FA), Objects (OB), RI birds

(RB), and Asian birds (AB) under 4 tasks.

•  Either PV or 1bID as FFA localizer task (p=.

05 corrected threshold)

-  Neuroelf for family-wise correction on voxel-

wise correlation

-  Custom matlab script for random-effect

voxel-wise partial correlation

type Sutdy fMRI

design Behavior

al task Expert domain (2) exp task

(3) Were stimuli in expert's specific expertise domain?

Behavioral

index (1) FFA localizer task

FFA activitiy measure (PSC:

percent signal change)

r (*: p<.05;

**: p<.01)

supportin g expertise hypothesi

s?

experts of natural type

Gauthie r et al

(2000) blocked sequentia l matching

car (n=6) 1-back location

yes

d'(car)-d'(bird)

passive viewing

PSC(car)-

PSC(bird) 0.75*

1-back identity 0.1 yes

bird (n=6) 1-back location d'(bird)-d'(car) PSC(bird)-

PSC(car) 0.82*

1-back identity 0.004

Xu(200

5) event- related

sequentia l matching

car (n=5)

left/right location

judgment yes

d'(car)-d'(bird)

1-back identity

PSC(car)-

PSC(bird) 0.41

yes

bird (n=5) d'(bird)-d'(car) PSC(bird)-

PSC(car) ** -0.91

car + bird combined (n=10)

d'(bird)-d'(car) PSC(bird)-

PSC(car) 0.61

d'(car) PSC(car)-

PSC(obj) 0.61

d'(bird) PSC(face)-

PSC(bird) -0.42

Rhodes et al

(2004) blocked classify butterflies

/moles

butterfly/ moths (n=8)

memory recognition

(individuation) no classification

accuracy passive viewing

PSC(old/new buttfly

recognition) 0.8**

no PSC(old/new obj

recognition) 0.84**

Grill- Spector

et al (2004)

event- related

sequentia l matching

car experts +

novices (n=10) detection/iden- tification (detect

cars/identify jeep)

partially

no d'(car) passive viewing

PSC(geep identified)-

PSC(car detected)

0.22

no car novices

(n=5) p<.03*

experts of artificial stimuli

Gauthie r et al.

(2002) blocked

half- composit e discrimin

ation

Greeble experts(n=5)

sequential matching of Greebles

no (untrained

)

composite - original effect (Δ

d') passive viewing Summed t values for object

of expertise p<.02* yes

Moore et al.

(2006)

event- related

simultane ous matching to sample

Polygon experts (n=11)

delayed sequential matching (working memory)

no both d'(WM) and d'(matching to

sample) 1-back identity

PSC(delayed period)experts - PSC (delayed period)novices

insignificant no

Op de Beeck et al.

(2006)

blocked shape discrimin

ation

Smoothies/

Spikies/Cubies experts (n=9)

a demanding color change

detection task n.a. discrimination

index n.a.

PSC(trained obj post-train)- PSC(untrained obj post-train)]- [PSC(trained obj

pretrain)- PSC(untrained

obj pretrain)]

p<.05 no

Wong et al.

(2009) blocked configural matching Ziggerin experts (n=18)

1-back basic/

subordinate discrimination

no ( untraine

d Ziggerins)

configural processing (match - mismatch RT)

1-back identity

PSC(post- training)- PSC(before_train ing)

within family:

r=.33, p=.177;

between: r=.43, p=.073

yes

(18)

Summary & Conclusions

Acknowledgement

Perceptual ExperHse Network (#15573-‐S6), a collaboraHve award from James S.

McDonnell FoundaHon, Taiwan NSC98-‐2410-‐H-‐006-‐002-‐MY2 to CCK, NSF award to

MJT: #BCS-‐0094491, and the Ihleson Family FoundaHon. .

!

Figure 6. The results of random-‐eﬀect voxel-‐wise parHal

correlaHon, or RVPC, analysis results. Each color denotes

contrast-‐speciﬁc parHal correlaHons across all brain voxels.

AcHvated clusters are above threshold of 8 conHguous voxels.

The [RB vs. OB]-‐correlated areas in 1bLO task was not shown.

Figure 5. LeU: the individual FFA map (N=17) under the PV

localizer task; Middle: the individual FFA map under the 1bID

localizer task; Right: the overlap between the overall FFA by

the PV task (orange color) and by the 1bID task (blue). The

top row represents both coronal and axial views, the middle

row the sagihal view, and the bohom row the glass brain

view.

Figure 8. The proposed mountainous ﬁgure aiming to explain the current

results: when the FFA was deﬁned by [Faces vs. Objects]@p.05 threshold in the

PV task, only the face-‐selecHve voxels are selected, thus reducing the likelihood

of genng experHse-‐selecHve voxels into the FFA-‐experHse correlaHon (thus non-‐

signiﬁcant); but when the same contrast was applied in 1bID task, the larger FFA

cluster include more experHse-‐related voxels into the correlaHon, thus rendering

the resulHng FFA-‐experHse correlaHon signiﬁcant. Further support also came

from the marginal signiﬁcant correlaHon results when p=.20 (and .40) under PV

task.

(a) the tentaHve mountainous representaHon in the

FFA both provides the possible reason behind

earlier null, and may combine with opHmal FFA

volume and localizer-‐experiment similarity as a

general requirement of signiﬁcant FFA-‐experHse

correlaHons;

(b) the complementary RVPC analyses addresses some

of the subtleHes underlying the nature of mFG,

which is modulated at least, but not limited, by task

demand, sHmulus familiarity, and experHse. More

speciﬁcally, the level of cogniHve load in a given

task aﬀects the category breadth of experHse-‐

predicated category-‐selecHve regions.

(c)  Other regions-‐of-‐interest besides the mFG, such as

ACC and PCC, may underlie the experHse processing

network as well

!

Figure 7. The overlapped experHse-‐predicated regions

between PV (green) and 1bID (blue) tasks, showing

both anterior and posterior cingulate cortex area (ACC

and PCC), along with bilateral mFG regions (marked by

red circles).

(d) Complementary analysis by random-‐

eﬀect voxel-‐wise par9al correla9on (RVPC)

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

1bID 1bLO 2bID

R² = 0.25905

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

PV 1bLO 2bID

PV_as_FFA_localizer

1bID_as_FFA_localizer

(e) ACC and PCC, along with bilateral mFG,

are among the default exper9se processing

networks

A simple explanatory framework

(c) FFA-‐exper9se correla9ons

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

PV 1bLO 2bID R² = 0.22

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

1bID 1bLO 2bID

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

1bID 1bLO 2bID

-‐0.4 -‐0.2 0 0.2 0.4 0.6 0.8

0 0.5 1 1.5 2 2.5

PV 1bLO 2bID

* Similar paHerns with RVC

Results suggest an “op9mal” FFA

volume, as well as the localizer-‐exp_task

similarity, in determining the FFA-‐

exper9se correla9onal signiﬁcance

References

1. Gauthier I, Tarr MJ. (2002): Unraveling mechanisms for expert object recogniHon: bridging brain acHvity and behavior. J Exp Psychol Hum

Percept Perform 28(2):431-‐46.

2. Gauthier I, Skudlarski P, Gore JC, Anderson AW. (2000): ExperHse for cars and birds recruits brain areas involved in face recogniHon. Nat

Neurosci 3(2):191-‐7.

3. Grill-‐Spector K, Knouf N, Kanwisher N. (2004): The fusiform face area subserves face percepHon, not generic within-‐category idenHﬁcaHon.

Nat Neurosci 7(5):555-‐62.

4. Moore CD, Cohen MX, Ranganath C. (2006): Neural mechanisms of expert skills in visual working memory. J Neurosci 26(43):11187-‐96.

5. Op de Beeck HP, Baker CI, DiCarlo JJ, Kanwisher NG. (2006): DiscriminaHon training alters object representaHons in human extrastriate

cortex. J Neurosci 26(50):13025-‐36.

6. Rhodes G, Byah G, Michie PT, Puce A. (2004): Is the fusiform face area specialized for faces, individuaHon, or expert individuaHon? J Cogn

Neurosci 16(2):189-‐203.

7. Wong AC, Palmeri TJ, Rogers BP, Gore JC, Gauthier I. (2009): Beyond shape: how you learn about objects aﬀects how they are represented in

visual cortex. PLoS ONE 4(12):e8405.

8. Xu Y. (2005): RevisiHng the role of the fusiform face area in visual experHse. Cereb Cortex 15(8):1234-‐42.

(19)

國科會補助計畫衍生研發成果推廣資料表

日期:2011/10/31

國科會補助計畫

計畫名稱: 側膝核在特殊閱讀障礙兒童與成人所扮演的角色計畫主持人: 龔俊嘉

計畫編號: 99-2410-H-006-044- 學門領域: 實驗及認知心理學

無研發成果推廣資料

(20)

99 年度專題研究計畫研究成果彙整表

計畫主持人：龔俊嘉計畫編號：99-2410-H-006-044- 計畫名稱：側膝核在特殊閱讀障礙兒童與成人所扮演的角色

量化

成果項目 ^{實際已達成}

數（被接受或已發表）

預期總達成數(含實際已

達成數)

本計畫實際貢獻百

分比

單位

備註（質化說明：如數個計畫共同成果、成果列為該期刊之封面故事 ...

等）

期刊論文 0 0 100%

研究報告/技術報告 0 0 100%

研討會論文 4 4 100%

論文著作篇

專書 0 0 100%

申請中件數 0 0 100%

專利已獲得件數 0 0 100% 件

件數 0 0 100% 件

技術移轉

權利金 0 0 100% 千元

碩士生 2 0 100%

博士生 0 0 100%

博士後研究員 0 0 100%

國內

參與計畫人力

（本國籍）

專任助理 1 0 100%

人次

期刊論文 0 0 100%

研究報告/技術報告 0 0 100%

研討會論文 0 0 100%

論文著作篇

專書 0 0 100% 章/本

申請中件數 0 0 100%

專利已獲得件數 0 0 100% 件

件數 0 0 100% 件

技術移轉

權利金 0 0 100% 千元

碩士生 0 0 100%

博士生 0 0 100%

博士後研究員 0 0 100%

國外

參與計畫人力

（外國籍）

專任助理 0 0 100%

人次

(21)

其他成果

(無法以量化表達之成

果如辦理學術活動、獲得獎項、重要國際合作、研究成果國際影響力及其他協助產業技術發展之具體效益事項等，請以文字敘述填列。)

無

成果項目量化 名稱或內容性質簡述

測驗工具(含質性與量性) 0

課程/模組 0

電腦及網路系統或工具 0

教材 0

舉辦之活動/競賽 0

研討會/工作坊 0

電子報、網站 0

科教處計畫加填項

目計畫成果推廣之參與（閱聽）人數 0

(22)

國科會補助專題研究計畫成果報告自評表

請就研究內容與原計畫相符程度、達成預期目標情況、研究成果之學術或應用價值（簡要敘述成果所代表之意義、價值、影響或進一步發展之可能性）、是否適合在學術期刊發表或申請專利、主要發現或其他有關價值等，作一綜合評估。

1. 請就研究內容與原計畫相符程度、達成預期目標情況作一綜合評估

■達成目標

□未達成目標（請說明，以 100 字為限）

□實驗失敗

□因故實驗中斷

□其他原因說明：

2. 研究成果在學術期刊發表或申請專利等情形：

論文：□已發表 □未發表之文稿 ■撰寫中 □無專利：□已獲得 □申請中 ■無

技轉：□已技轉 □洽談中 ■無其他：（以 100 字為限）

3. 請依學術成就、技術創新、社會影響等方面，評估研究成果之學術或應用價值（簡要敘述成果所代表之意義、價值、影響或進一步發展之可能性）（以 500 字為限）

由於僅獲得一年之補助，本研究原本預計之實驗需要先由行為典範與 fMRI 實驗典範分頭進行。而兩者都獲得了相當的成果。學生林源欽將行為實驗部分合併大專生國科會計畫，

並順利施測，參加台灣心理學年會並報告壁報論文，目前正在撰寫初稿中。fMRI 部分，

由碩士生 Mahen 進行的兩地(包括在高醫與中國醫分別進行之正常人)資料蒐集與分析，

已在香港視覺年會中報告，並準備寫成碩士論文並投稿。未來將利用此兩方面進行之成果，搭配未來在成大 3T MRI 中心的設立，繼續進行閱讀困難孩童與成人的蒐集與施測。

側膝核在特殊閱讀障礙兒童與成人所扮演的角色

行政院國家科學委員會專題研究計畫 成果報告