DSpace community: 生醫理工學院

運動、學習與大腦：行為可訓練性、腦部結構形態與真實情境神經動態;Sport， Learning， and the Brain: Behavioural Trainability， Structural Morphology， and Real-World Neural Dynamics

Tue, 20 Oct 2026 09:01:04 GMT

title: 運動、學習與大腦：行為可訓練性、腦部結構形態與真實情境神經動態;Sport， Learning， and the Brain: Behavioural Trainability， Structural Morphology， and Real-World Neural Dynamics abstract: 運動與身體活動經常被認為能促進認知功能、行為表現及腦部健康，然而此類效益並非在所有認知作業、運動類型、神經指標或時間尺度上皆一致。本論文從三個互補層面探討運動相關的認知與腦部差異，包括行為學習、結構性磁振造影（MRI）及真實運動情境下的腦電圖（EEG）。本研究的核心目的並非單純判斷運動對大腦具有益處或害處，而是探討在不同認知作業、解剖尺度及實際運動情境下，運動相關差異如何表現於行為與神經層面。首先，本研究探討運動相關的認知優勢究竟反應較高的基礎能力，或是較佳的學習可塑性。在隱性動作序列學習作業中，比較田徑、棒球選手與缺乏運動習慣者的表現，結果顯示運動員具有較快的基礎動作反應速度，但在一般動作技能學習及序列特定學習方面並未展現一致優勢。研究進一步利用情境提示（contextual cueing）視覺搜尋作業檢驗學習能力，雖然所有受試者皆呈現穩健的視覺搜尋學習效果，但運動參與程度及體適能並未與較高的學習速率相關。整體而言，研究結果顯示運動所帶來的認知優勢較偏向特定能力，而非支持「整體學習能力提升」的假說。其次，本研究利用多種互補性的MRI分析方法探討運動相關的腦部形態差異。在大學美式足球員、越野田徑選手與非運動員之間，比較體素形態分析（FSL-VBM）、皮質下結構分析（FIRST）及腦組織分割（FAST）結果，發現腦幹、皮質下結構及腦組織組成存在差異，但其型態依比較對象而有所不同，凸顯納入非碰撞性運動員作為對照組的重要性。另一項針對菁英足球選手的研究則發現局部灰質差異、廣泛的皮質下體積變化、局部形狀改變以及較低的絕對腦組織體積。然而，組織比例分析及雙向形狀變化結果顯示，這些差異不宜簡化解釋為廣泛性的神經退化。最後，本研究利用事件鎖定高密度EEG分析桌球實際擊球時的腦神經動態。相較於與發球機互動，受試者在真人對打情境中呈現不同的擊球前後頻譜活動。殘差頻帶功率分析顯示，發球機條件下的α波與β波功率較低；然而，頻譜參數化分析進一步指出，最穩定且明顯的差異為發球機情境下非週期性頻譜指數（aperiodic exponent）增加，且主要分布於後部及頂葉區域。綜合而言，本研究證實運動相關的認知與神經差異確實存在，但其表現並非單一且普遍一致。無論是在行為學習、結構性MRI或真實情境EEG分析中，研究結果皆支持運動效應具有高度特異性，而非普遍性的提升或損害。運動對大腦與認知的影響應視為訓練經驗、作業需求、解剖尺度、暴露特性、測量方法及即時運動情境等多重因素交互作用的結果。;Sport and physical activity are often associated with cognitive， behavioural， and neural benefits， but these associations are not uniform across tasks， sports， brain measures， or timescales. The present work examined sport-related cognition and brain differences across three complementary levels of analysis: behavioural learning， structural magnetic resonance imaging， and real-world electroencephalography. The central aim was not to determine whether sport is simply beneficial or harmful for the brain， but to ask how sport-related differences are expressed when cognition and neural measures are examined across different tasks， anatomical scales， and performance contexts. The first part of the work investigated whether sport-related cognitive advantages are better understood as altered baseline competence or enhanced trainability. In one study， runners， baseball players， and non-exercising controls completed an implicit motor sequence learning task that separated baseline response speed， general motor skill learning， and sequence-specific learning. Sport-trained participants showed evidence of faster baseline motor performance than sedentary controls， but did not show a consistent advantage in general motor skill learning or sequence-specific learning. A second behavioral study extended this question to visual search using a contextual cueing task. Although robust visual-search learning and contextual cueing were observed， sport and fitness were not associated with a broad increase in learning rate. Together， these findings suggest that sport-related behavioral advantages may be more specific than a general “more trainable brain” account would predict. The second part examined sport-related structural brain morphology using complementary MRI methods. In collegiate American football players， cross-country runners， and non-athletes， FSL-VBM， FIRST， and FAST analyses revealed brainstem， subcortical， and tissue-composition differences， but the pattern depended on the comparison group. This supported the value of including a non-collision athlete group when interpreting morphology is collision sports. In elite soccer players and matched controls， structural analyses identified focal grey-matter differences， widespread subcortical volumetric differences， local shape effects， and lower absolute tissue volumes. However， tissue-fraction and bidirectional shape findings caution against a simple interpretation of diffuse degeneration. The final part examined real-world neural dynamics during table tennis using event-locked high-density EEG. Human-player and ball-machine contexts were associated with systematic peri-hit spectral differences. Residualised bandpower suggested lower ball-machine than human-player alpha and beta power in several comparisons， but spectral parameterisation showed that the clearest context effect was a higher aperiodic exponent during ball-machine trials， especially over posterior and parietal regions. Overall， the findings indicate that sport-related cognitive and neural differences are real but not simple. Across behavioural learning， structural MRI， and real-world EEG， the evidence points towards specificity of effects rather than generality. Sport-related brain and cognition differences should therefore be understood as the product of interacting factors， including training history， task demands， anatomical scale， exposure profile， measurement method， and immediate performance context.

視覺工作記憶精準度與額葉 Theta 波及後頂葉 Alpha/Beta 振盪之非線性交互作用;Visual Working Memory Precision is Supported by Nonlinear Interactions Between Frontal Theta and Posterior-parietal Alpha/Beta Oscillations

Tue, 20 Oct 2026 09:01:01 GMT

title: 視覺工作記憶精準度與額葉 Theta 波及後頂葉 Alpha/Beta 振盪之非線性交互作用;Visual Working Memory Precision is Supported by Nonlinear Interactions Between Frontal Theta and Posterior-parietal Alpha/Beta Oscillations abstract: 本論文運用全息希爾伯特頻譜分析（Holo-Hilbert Spectral Analysis， HHSA）探討工作記憶精準度的腦波振盪基礎。HHSA 為一種適應性的雙層分解方法，能在不預設固定頻帶、亦不假設訊號呈正弦波形的前提下，量化非線性的跨頻率交互作用。33 名年輕成人執行連續回報的顏色作業，記憶負荷分為四種水準（1、2、4、6），同時記錄 64 通道腦電波圖（EEG）。回憶精準度以三成分混合模型所估計的集中度參數 κ 作為指標。試次平均的 HHSA 分析（第五章）顯示，後頂枕區 alpha/beta 呈現隨記憶負荷遞增的抑制。額葉中線 theta 則未顯示顯著的負載量效果。在個別差異層次上，較高的額葉 theta 與較低的後側 alpha/beta 分別與較高的 κ 共變。回溯提示出現後，後側 alpha/beta 包絡的第二層振幅調變（amplitude modulation， AM）訊號亦與 κ 共變。額葉 theta 對後側 alpha/beta 的相位－振幅耦合隨記憶負荷增加而增強，且以額葉 theta 相位領先，但未能預測 κ。後續的源定位分析（第七章）將第五章的各項神經相關指標定位至大腦皮質層次。額葉 theta 定位於內側前額葉皮質／背側前扣帶皮質（mPFC/dACC）及背內側前額葉皮質（dmPFC）等分區，後側 alpha/beta 則對應至視覺與頂葉皮質。在全腦層次上，僅 IMF4（beta）於後頂枕皮質呈現顯著的 κ 負向精準度叢集，IMF5（alpha）則未達顯著。單一試次分析（第六章）進一步檢驗上述特徵。受提示半視野的空間注意力解碼器於 +1.37 秒達到峰值（AUC = 0.569，p < .0005），其特徵重要性落於對側後側 alpha。精準度解碼器在第一層包絡上未達顯著；僅於後側 alpha 與 beta 接近趨勢水準的振幅變化達到顯著（峰值 r = 0.067，位於 +1.83 秒；叢集校正 p = .0065）。一旦納入後側叢集，額葉特徵即不再攜帶獨立的精準度訊號，顯示額葉的貢獻僅限於協調功能。顏色內容編碼則僅於反應階段出現，在記憶維持階段並未發現。上述發現支持一個三成分的解釋架構。後側 alpha/beta 包絡作為感覺內容的增益調節器，其中 beta 追蹤回憶保真度，alpha 追蹤記憶負荷；額葉中線 theta 為執行控制的組織節律；跨頻率耦合則為一條協調通道，其強度隨負載量增加，但並不決定試次層次的精準度。本論文提出一套源定位、以 HHSA 為基礎的時間延遲嵌入式隱馬可夫模型（time-delay embedded hidden Markov model， TDE-HMM），作為未來工作及分析的共同架構。;This thesis examined the oscillatory basis of working memory (WM) precision using Holo-Hilbert Spectral Analysis (HHSA)， an adaptive two-layer decomposition that quantifies nonlinear cross-frequency interactions without fixed bands or sinusoidal assumptions. Thirty-three young adults performed a continuous-report color task at four set sizes (1， 2， 4， 6) with a target probe while 64-channel EEG was recorded. Recall precision was indexed by κ from a three-component mixture model. Trial-averaged HHSA (Chapter 5) showed load-graded parieto-occipital alpha/beta suppression. Frontal midline theta showed no significant load effect. Across individuals， higher frontal theta and lower posterior alpha/beta each covaried with higher κ. A post-retrocue second-layer amplitude-modulation (AM) signal in the posterior alpha/beta envelope also covaried with κ. Frontal-theta-to-posterior-alpha/beta phase-amplitude coupling scaled with set size， with frontal theta phase leading， but did not predict κ. Subsequent source analysis (Chapter 7) localized the Chapter 5 correlates to cortex. Frontal theta was localized to mPFC/dACC and dmPFC parcels， and posterior alpha/beta to visual and parietal cortices. At the whole-brain level， only IMF4 (beta)， not IMF5 (alpha)， carried a significant κ-negative precision cluster over parieto-occipital cortex. Single-trial decoding (Chapter 6) tested these features. A cued-hemifield spatial-attention decoder peaked at +1.37 s (AUC = 0.569， p < .0005)， with importance in contralateral posterior alpha. The precision decoder was null for the first-layer envelope. Significant was only found for the near-trend envelope component of posterior alpha and beta (peak r = 0.067 at +1.83 s; cluster-corrected p = .0065). Frontal features carried no independent precision signal once posterior clusters were included， confining the frontal contribution to coordination. A color-identity code emerged only at the response stage. These findings support a three-component account. Posterior alpha/beta envelopes act as a gain modulator on sensory content， beta tracking fidelity and alpha tracking load. Frontal midline theta is the executive organizing rhythm. Cross-frequency coupling is a coordination channel that scaled with load but did not determine trial-level precision. A HHSA-informed time-delay embedded hidden Markov model (TDE-HMM) is proposed as the main analysis framework for future works.

視網膜電圖訊號分析：運用全希爾伯特頻譜分析於不同遺傳性視網膜疾病;Electroretinography Signal Analysis Of Different Inherited Retinal Diseases Using Holo-Hilbert Spectrum Analysis

Tue, 20 Oct 2026 09:00:59 GMT

title: 視網膜電圖訊號分析：運用全希爾伯特頻譜分析於不同遺傳性視網膜疾病;Electroretinography Signal Analysis Of Different Inherited Retinal Diseases Using Holo-Hilbert Spectrum Analysis abstract: 遺傳性視網膜疾病（inherited retinal diseases， IRDs）在臨床表現與基因型上皆具高度異質性，而病灶的空間分布往往決定了全視野視網膜電圖（full-field electroretinogram， ffERG）能否偵測到病變。廣泛性退化會反映在整體的全視野訊號中，局部性的黃斑部失養症（macular dystrophy， MD）卻常被掩蓋，因為病變的黃斑部僅佔整體反應的一小部分，而此反應主要由健康的周邊視網膜所主導。本研究旨在探討：ffERG 傳統上以能量為基礎的判讀所遺漏的訊息，是否能藉由非線性、非定態的調變分析加以還原，並特別著重於黃斑部失養症。本研究使用取自國立臺灣大學醫學院附設醫院門診病患、經去識別化的 ffERG 臨床紀錄，共 174 位病患、188 筆檢查，涵蓋七個臨床分組：健康對照組、黃斑部失養症、熄滅型（flat）ERG、錐桿細胞失養症、Bietti 結晶樣視網膜失養症，以及典型與中心周圍型色素性視網膜炎。每筆反應皆在五種 ISCEV 標準刺激條件下記錄，並以改良式完整集成經驗模態分解（ICEEMDAN）進行分解，再以希爾伯特—黃轉換（HHT）、振幅與頻率調變之全希爾伯特頻譜分析（HHSA-AM、HHSA-FM）及內在事件相關電位（iERP）加以分析。組間差異以群集式置換檢定評估，並以兩項經振幅正規化的個體層級指標整合上述結果圖。此外，透過僅保留每位病患單一紀錄的敏感度分析，以及納入年齡為共變量的共變數分析（ANCOVA），檢驗結果的穩健性。結果顯示，各廣泛性退化組在所有條件下皆與對照組明顯區隔，證實了分析流程的有效性；相對地，黃斑部失養症在以能量為基礎的 HHT 中幾乎無法與對照組區分，僅在中間層、內層視網膜相關的條件下達到顯著。調變域則還原出一組雖小卻一致、且侷限於錐細胞路徑的黃斑部失養症特徵，其中最具辨識度者，為錐細胞單獨刺激下的振幅調變過度增強——此方向在所有組別中獨一無二，其餘各組皆呈下降。一項跨條件的錐細胞區功率比值，使黃斑部失養症成為唯一高於健康基準的分組，並經單因子變異數分析證實。該效應對於世代中重複回診的資料結構及受試者年齡皆具穩健性；由於黃斑部失養症為最年輕的;Inherited retinal diseases (IRDs) are clinically and genetically heterogeneous disorders in which the spatial distribution of pathology determines whether the full-field electroretinogram (ffERG) can detect them. Generalized degenerations are expressed in the combined full-field signal， whereas focal macular dystrophy (MD) is characteristically concealed， because the affected macula contributes only a small fraction of a response dominated by the healthy peripheral retina. The purpose of this study was to test whether information overlooked by the conventional energy-based reading of the ffERG can be recovered through nonlinear， non-stationary modulation analysis， with particular attention to macular dystrophy. The data comprised de-identified clinical ffERG recordings obtained from outpatients at National Taiwan University Hospital， totaling 188 examinations from 174 patients across seven clinically defined groups: healthy control， macular dystrophy， extinguished or flat ERG， cone-rod dystrophy， Bietti crystalline dystrophy， and typical and pericentral retinitis pigmentosa. Each response was recorded under five ISCEV-style stimulus conditions and decomposed using improved complete ensemble empirical mode decomposition with adaptive noise. The signals were then analyzed with the Hilbert-Huang transform (HHT)， Holo-Hilbert spectral analysis of amplitude and frequency modulation (HHSA-AM and HHSA-FM)， and the intrinsic event-related potential (iERP). Group differences were assessed by cluster-based permutation testing， and two amplitude-normalized， subject-level indices condensed the resulting maps. Robustness was examined through a sensitivity analysis restricting the data to one recording per patient and through an analysis of covariance controlling for participant age. The generalized groups separated strongly from controls across all conditions， validating the analytic pipeline， whereas macular dystrophy was largely indistinguishable from controls in the energy-based HHT， reaching significance only in an intermediate-layer， inner-retinal condition. The modulation domain recovered a small but convergent macular-dystrophy signature confined to the cone pathway. Its most distinctive feature was an amplitude-modulation excess under cone-isolated stimulation， unique in sign among all groups， which were otherwise uniformly suppressed. A cross-condition cone-region power ratio placed macular dystrophy alone above the healthy baseline， and the group effect was confirmed by one-way analysis of variance. The effect was robust to the repeat-visit structure of the cohort and to participant age; because macular dystrophy was the youngest group， the age association was conservative with respect to the finding. The effect further localized to a single cone-dominant intrinsic mode. These findings indicate that the full-field electroretinogram carries recoverable， macular-relevant information in its modulation structure that a conventional energy reading discards. Holo-Hilbert analysis can therefore convert a routine ffERG into a more sensitive probe of retinal response organization， supporting a candidate modulation-domain signature for macular dystrophy that the present cross-sectional design identifies but cannot yet validate for individual diagnosis.

探討視覺知覺學習與基於錯誤的動作學習之間的關聯;Exploring the Connections Between Visual Perceptual Learning and Error-Based Motor Learning

Tue, 20 Oct 2026 09:00:50 GMT

title: 探討視覺知覺學習與基於錯誤的動作學習之間的關聯;Exploring the Connections Between Visual Perceptual Learning and Error-Based Motor Learning abstract: 本研究旨在探討視覺知覺與動作兩個領域的學習機制，究竟是由領域共通（domain-general）還是領域特定（domain-specific）的機制所調控。儘管先前研究已發現這兩個系統在學習動態上具有相似性，知覺學習與動作學習之間直接的神經與行為關聯仍不明確。為釐清視覺知覺學習與基於錯誤的動作學習之間的關聯，我們以紋理辨識任務（TDT）評估視覺知覺學習，並以視覺動作適應任務（VAT）評估基於錯誤的動作學習。實驗一檢驗是否存在一個共通的學習機制，以支持跨領域的學習進步速度（學習速率）。透過非線性指數衰退模型捕捉動作適應的快速動態，我們發現在所有階段中，包括早期適應期（r = -.21)、晚期適應期（r = -.15）與再適應期（r = .002）TDT與VAT 的學習表現之間皆無顯著相關。此一行為結果顯示，個體在視覺系統中快速可塑性的能力，未必對應其在基於錯誤之動作校正上的效率，這很可能反映兩者所涉及的神經機制有別，例如TDT 主要仰賴早期視覺皮層，而VAT則涉及小腦的處理。實驗二則是建立在行為層次彼此獨立的基礎上，進一步探討更複雜的指標，包括學習轉移與靜息態連結性。結合靜息態連結性、任務相關腦電圖（EEG）以及格蘭傑因果關係（GC）的分析，揭示了一種階層性的解離現象（hierarchical dissociation）。兩項學習任務皆誘發了共通的整體「神經效率」模式，其特徵為靜息態 alpha 頻段（8–12 Hz）與低頻 gamma 頻段（30–55 Hz）連結性的下降。然而，主動的處理路徑仍維持高度的領域特定性：視覺知覺學習呈現「整合」模式，表現為由後至前（posterior-to-anterior）傳向額葉—中央區域的資訊流增加；動作學習則遵循「自動化」模式。兩領域間更深層的連結，則反映於 TDT 轉移指數與動作再適應率之間的顯著相關（r = -.57），其中 theta 頻段（4–8 Hz）可能作為錯誤監控的領域共通控制中樞。綜上所述，本研究結果顯示，儘管以行為進步速率衡量的學習能力屬於領域特定的歷程，但其底層的神經可塑性與錯誤監控系統，則是透過跨越視覺與動作領域的領域共通架構而運作。;This study investigates whether learning across the visual perceptual and motor domains is governed by domain-general or domain-specific mechanisms. Although prior research has documented similarities in the learning dynamics of these two systems， the direct neural and behavioral relationship between perceptual and motor learning remains unclear. To clarify the link between visual perceptual learning and error-based motor learning， we employed the Texture Discrimination Task (TDT) to assess visual perceptual learning and the VAT to assess error-based motor learning. In Experiment 1 (N = 51)， we examined whether a shared mechanism supports the rate of improvement (learning rate) across the two domains. Using non-linear exponential decay modeling to capture the rapid dynamics of motor adaptation， we found no significant correlation between TDT and VAT performance in any phase， including the early adaptation (r = -.21)， late adaptation (r = -.15)， and de-adaptation (r = .002) phases. These behavioral findings indicate that an individual’s capacity for rapid plasticity in the visual system does not necessarily correspond to their efficiency in error-based motor correction. This likely reflects the distinct neural substrates involved， namely the early visual cortex for the TDT and cerebellar processing for the VAT. Experiment 2 (N = 26) built upon this baseline of behavioral independence by examining more complex measures， including learning transfer and resting-state connectivity. Analyses combining resting-state connectivity， task-based EEG， and Granger causality (GC) revealed a pattern consistent with a hierarchical dissociation. Both learning tasks induced a shared global pattern of neural efficiency， characterized by reduced resting-state alpha (8–12 Hz) and low-gamma (30–55 Hz) connectivity. In contrast， the active processing pathways remained highly domain-specific. Visual perceptual learning exhibited an integration pattern， reflecting increased posterior-to-anterior information flow toward the fronto-central region， whereas motor learning followed an automation pattern. A deeper cross-domain link was evidenced by a significant correlation between the TDT transfer index and the motor de-adaptation rate (r = -.57)， with the theta band (4–8 Hz) potentially serving as a shared， domain-general control hub for error monitoring. In sum， our findings suggest that although learning capacity， as measured by the behavioral rate of improvement， is a domain-specific process， the underlying neural flexibility and error-monitoring systems operate through a domain-general framework spanning the visual and motor domains.