預測處理功能於年輕與老年神經迴路之研究

Abstract

人類大腦反應外界訊息的過程可以貝氏推理的方法來呈現。這個方法是先利用既有的先驗信念來解讀資料，之後將概度整合以形塑成事後推論，而這兩個步驟是適應變化中環境的重要認知歷程。在這個研究中，我們假設年齡會改變大腦的神 經迴路，而這樣的變化使得年輕人與老年人以不同的能力來適應不斷變化的環境。 34 位年輕人(年齡 = 22.66 ± 2.58 歲，20 位女性)與 32 位老年人(年齡= 70.64 ± 5.60 歲， 17 位女性)參與我們的功能性磁振造影實驗。實驗任務會使用的三個箱子包含紅、藍、黃三種顏色的球，實驗參與者在實驗任務之前將會學習不同顏色的球的出現機率。在正式實驗任務中，三顆球將會從其中一個被選定的箱子中抽 出，而實驗參與者需要猜測待會抽出的三顆球是哪一個顏色比較多，之後會將抽出的三顆球呈現給參與者，並按照結果給予獎賞。參與者需結合自己的抽球前的先驗信念搭配抽球結果的概度來得出事後推論，才能準確判斷哪一個箱子才是被選定的箱子，而正確的推論能準確地預測抽球結果，並獲得更多的分數。實驗中，被選定的箱子將在一段時間之後更換為另一個箱子，所以參與者必須留心抽球結 果，並在準確地在正確的時間轉換先驗信念到另一個箱子。我們以貝氏修正 (Bayesian updating)的計算模型來適配參與者的行為資料，老年人相對於年輕人有比較低的機率保留原本的先驗信念(Cohen’s d = 0.86, 95%可靠區間 = [0.34, 1.38])，表示老年人的信念較不穩固，不顧抽球結果的概度就頻繁地更改心中對箱子的信念。而且，在整合抽球結果的概度上，並沒有發現統計上顯著的年齡差 異(Cohen’s d = -0.35, 95%可靠區間 = [-0.89, 0.20])。在功能性大腦影像資料中，我們因嚴重的頭動狀況排除一位年輕人與一位老年人，而剩餘的 33 位年輕人(年齡 = 22.72 ± 2.60 歲，19 位女性)與 31 位老年人(年齡 = 70.70 ± 5.68 歲，17 位 女性)以表徵相似分析(representational similarity analysis，RSA)與模型為基的單 變量廣義線性模型(model-based univariate generalized linear model)進行分析。 相對於年輕人，老年人功能性影像的右側頂下葉(inferior parietal lobe; 54, -42, 51)與楔前葉(precuneus; 0, -48, 45)的多變項神經模式(multivariate neural patterns)與後驗信念有比較差的關聯性，表示信念在大腦的神經表徵與老年人的 決策行為有不一致的狀況。另外，年輕人在腹內側前額葉(ventral medial prefrontal cortex; 0, 48, -15)與海馬迴(hippocampus; -33, -30, -15)的活化程度 與信念的信心有正相關，但在老年人的大腦上沒有這樣的正相關。總體來說，我們在行為上發現老年人的信念較不穩定，而這能反映在頂下葉與楔前葉有比較差 的神經表徵。相對應地，老年人減少腹內側前額葉與海馬迴的活化來維持信念的 信心。另外，在本論文中，我們結合預測性編碼(predictive coding)的理論架構來建立假說，我們預測大腦在低階層腦區或高階層腦區若遭受影響，將分別導致信念修正上比較頑固死板或比較反覆無常。我們的發現支持後者的假設，老年人後側頂葉的神經處理的失序使得在表徵高層次的抽象信念遇到問題，反映在決策行為上，與我們觀察到老年人有不穩定信念的結果一致。

Human brains respond to incoming external information in a manner that can be formulated as Bayesian inference. Exploiting internal prior beliefs to make inferences about incoming information and then integrating likelihoods to form posterior inferences are two critical cognitive processes facilitating compliance with the changing environment. We postulated that age alters the neural circuits that reflect different abilities to adapt to changing environments in younger and older adults. In a functional magnetic resonance imaging (fMRI) experiment, we recruited 34 younger adults (age = 22.66 ± 2.58, 20 female) and 32 older adults (age = 70.64 ± 5.60, 17 female). Participants first learned about different proportions of red, blue, or yellow balls contained within three boxes. At test, participants were told that sets of three balls would be drawn from one of the boxes and they had to guess the majority color of the balls drawn with the source box hidden from participants. Outcomes of the drawn ball colors were then provided with correct guesses rewarded. Thus, participants had to update their posterior beliefs about which was the source box by integrating prior beliefs about the box and the likelihood of the drawn colors. Critically, source boxes were intermittently switched so that participants had to notice the outcome changes and transition to new beliefs. Computational model fitting with Bayesian updating showed older adults had lower probabilities to stay in the same belief (Cohen’s d = 0.86, 95% credible interval: [0.34, 1.38]), indicating less stable beliefs that frequently transitioned from trial to trial regardless of outcome likelihoods. However, we found no evidence of age-related difference in updating beliefs from outcome likelihoods (Cohen’s d = -0.35, 95% credible interval = [-0.89, 0.20]). We excluded one younger adult and one older adult for the functional brain data because of severe head motion during scanning. Functional brain imaging from the remaining 33 younger adults (age = 22.72 ± 2.60, 19 female) and 31 older adults (age = 70.70 ± 5.68, 17 female) were submitted to representational similarity analysis (RSA) and model-based univariate generalized linear modeling. Compared to younger adults, older adults showed reduced association between multivariate patterns and trial-wise posterior beliefs in the right inferior parietal lobe (IPL; 54, -42, 51) and precuneus (0, -48, 45), suggesting a behavioral disjoint with neural representations of belief in older adults. In younger adults, belief confidence also modulated activation in ventral medial prefrontal cortex (vmPFC; 0, 48, -15) and hippocampus (-33, -30, -15), which was not seen in older adults. Overall, we show that older adults have less stable beliefs. This is reflected by reduced neural representations of posterior beliefs in IPL and precuneus in older. Correspondingly, older adults showed reduced recruitment of VMPFC and hippocampus to maintain confident beliefs. Furthermore, in this dissertation, we incorporated a theoretical framework based on predictive coding and hypothesized that the disruption in brain regions at lower-level or higher-level hierarchies results in rigid or unstable beliefs, respectively. Our findings support the later speculation that older adults have disordered posterior parietal cortex (PPC) engagement in representing higher-level abstract beliefs which is associated with stochastic beliefs exemplified in decision behavior.