人腦對於聽覺序列訊息的預測誤差訊號之階層性神經機制: 事件相關腦電位研究

Abstract

預測是大腦的重要功能之一，能利於個體成功地應對環境變化。預測編碼理論是近期被視為能廣泛解釋大腦各種作用與現象的優勢理論。此理論認為大腦內預測訊號的產生與更新是一個雙向的階層性訊息處理過程：大腦一般會根據過去經驗產生一個由上而下傳遞的預測訊號；當外在之訊息進入大腦時，若外在訊息不符預測，預測誤差訊號會產生並由下而上傳遞，進而修正預測。預測與預測誤差之階層性交互關係為其理論之核心概念，然而其神經證據仍尚未被系統性地驗證。本研究著重探究人腦解讀聽覺序列訊息時，預測訊號與預測誤差訊號之階層性交互關係的神經機制。實驗設計採用新改良版的局部－整體聽覺刺激，透過操弄兩階層間之規則來建立不同強度的預測訊號強度：局部預測規則（單一刺激呈現頻率）與整體預測規則（序列刺激呈現頻率）。由腦電圖量測的預測誤差訊號之特性，能藉由分析違反局部預測規則之特異刺激與標準刺激所產生的腦波反應差異。本研究分別以事件相關電位了解階層性預測誤差處理之歷程： Mismatch negativity （MMN） 代表局部預測誤差訊號； P300 代表整體預測錯誤訊號。研究結果顯示：（一）MMN高峰振幅主要以單一刺激呈現頻率解釋兩者負關係；而延遲時間亦由單一刺激呈現頻率解釋兩者正關係。（二）P300高峰振幅主要由單一和序列刺激呈現頻率解釋其正相關；而延遲時間則由序列刺激呈現頻率解釋兩者正關係。（三）局部預測誤差訊號在時序上，先隨單一刺激呈現頻率而變化，後隨單一與序列刺激呈現頻率而變化。事件相關電位之調節結果表示，預測誤差處理於較低階層時受局部預測規則影響；而預測誤差訊號由下而上傳遞之較高階層時，其誤差由整體預測規則做解釋。此預測錯誤訊號之階層性神經證據將有助於佐證預測編碼之重要概念。

Human brains are thought to be a prediction machine enabling us to be adaptive to various environmental changes through minimizing adjusting predictions and minimizing prediction errorsor adjusting prediction. According to the predictive coding theory, a leading theory of brains, postulates such processes are achieved by the bidirectional cascades of cortical processes: predictions are constantly sent from higher to lower cortical areas to match incoming sensory inputs, and prediction errors are sent from lower to higher cortical areas to update predictions when a mismatch occurs. Neural interactions between these two streams of signals across hierarchies are the core hypothesis of the theory; however, their empirical evidence has not been systematically examined. This study aimed to investigate the neural interactions between prediction and prediction-error signals across two hierarchies during the processes of auditory sequences. We adopted the novel extension of the local-global auditory oddball paradigm in which strengths of predictions were established by the local regularity (the tone probability) and the global regularity (the sequence probability). Prediction-error signals from the EEG recording were measured by comparing neural responses to oddballs with standard stimuli. The two event-related potential (ERP) components were used to examine hierarchical prediction-error processes: mismatch negativity (MMN) indexed as local prediction-error signals, and P300 indexed as global prediction-error signals. There were several important findings. First, for the MMN components, the tone probability shows great significant contribution (negative effect) to the peak amplitudes, and also significant contribution (positive effect) to the latencies. Secondly, for the P300 component, both tone and sequence probabilities show great significant contribution (positive effect) to the peak amplitudes, whereas the sequence probability shows significant contribution ( positive effect) to the latencies. Third, the neural responses of prediction-error signals temporally changed with the tone probability in the early stage and with both probabilities in the late stage. In conclusion, the modulations of the ERP responses suggested that lower-level prediction-error processes may be influenced by the local regularity, and at a global level, the bottom-up error signals may be further explained away with the help of global regularity. This evidence of the neural representations of prediction-error signals across two hierarchical levels of regularities can support the concept of the hierarchical predictive coding theory.