- 學位論文
以相互資訊和相關係數所計算之中腦與大腦靜息態功能性連結差異
Mutual Information and Correlations Capture Different Midbrain-Cerebrum Resting-State Functional Connectivity
摘要
過往學術上分析腦部在靜息態 (Resting-State) 下的功能性連結 (Functional Connectivity) 時經常使用皮爾森相關係數 (Pearson Correlation Coefficient) 來描述腦區間訊號呈現線性關聯之程度。然而，腦內的神經訊號傳遞亦可能具有非線性之關聯。作為量化資料間關聯的方式，相互資訊 (Mutual Information, MI) 不需仰賴預設模型，因此亦具備捕捉非線性關聯之潛力。本研究探討使用MI量化大腦功能性連結之可能性，並聚焦於中腦分泌單胺類神經傳導物質的四個神經核至全腦之連結。研究中分析了135位20-28歲健康受試者的靜息態功能性磁振造影結果，並分別利用相關係數及MI計算中腦分泌多巴胺的腹側背蓋區 (Ventral Tegmental Area, VTA) 、分泌血清素的背側縫核 (Dorsal Raphe Nuclei, DRN) 、中央縫核 (Median Raphe Nuclei, MRN) ，以及分泌正腎上腺素的藍斑核 (Locus Coeruleus, LC) 等四個種子區域對大腦的功能性連結。 使用相關係數運算時，我們發現VTA與前扣帶皮層 (Anterior Cingulate Cortex) 、視丘 (Thalamus) 和舌回 (Lingual Gyrus) 等區域具有較顯著的功能性連結，而DRN、MRN及LC等區域則與視丘、枕葉和小腦有顯著的功能性連結。使用MI所算出的功能性連結圖譜大致上與使用相關係數所算出之圖譜相符，代表MI至少能夠捕捉到訊號間的線性關聯。我們更進一步將兩方法運算出的圖譜相減，發現MI偵測到更多中腦與額葉 (Frontal Lobe) 的功能性連結。相對的，相關係數偵測到的功能性連結則多與視丘、大腦後側 (Posterior Brain) 及小腦相關。這些結果顯示MI至少反映了腦內的線性功能性連結，而MI亦具有潛力成為衡量靜息態磁振造影下訊號間複雜關聯性的一個附加分析工具。
並列摘要
Correlation coefficients, the most common metric of functional connectivity (FC) in resting-state functional magnetic resonance imaging (rs-fMRI), only characterize linear associations between voxels. Communication in neural networks, however, arguably comprises non-linear signals as well. Mutual information (MI) is a model-free approach sensitive to non-linear associations. We explored the feasibility of using MI to quantify FC, focusing on midbrain monoaminergic nuclei with their widespread efferents to the whole brain. In this study, rs-fMRI data of 135 young adults between ages 20-28 were analyzed. Four midbrain regions-of-interest (ROI) were chosen as seed regions including the ventral tegmental area (VTA), dorsal and median raphe nuclei (DRN, MRN), and locus coeruleus (LC), sources of dopamine, serotonin, and norepinephrine neurotransmitters. Functional connectivity was calculated from the four seed regions to the rest of the brain using correlations (FC⍴) and MI (FCMI). From the results of the analysis, VTA FC⍴ was observed mainly with anterior cingulate, thalamic, and lingual areas. DRN, MRN, and LC FC⍴ were observed mainly with thalamic, occipital, and cerebellar areas. FCMI broadly corresponded with these FC⍴ patterns, suggesting MI detects linear correlations, at least. Further, contrast maps of FCMI>FC⍴ suggest that FCMI captures more of the connectivity between the midbrain and the frontal regions, whereas contrasts of FC⍴>FCMI better capture midbrain connections with the thalamus, posterior brain, and cerebellar regions. Our findings demonstrate that MI can at least reflect linear correlations in inter-voxel FC and suggest that it is a plausible complementary tool to be used to evaluate more complex relationships that might be present in rs-fMRI data.