使用大腦區域-全域分割法探討臉部負面表情引起之腦部活動預測憂鬱症個體差異

Abstract

近年許多研究運用靜息態與作業相關功能性磁振造影（tfMRI）揭露腦部影像與憂鬱症的關聯。儘管如此，腦部活動與憂鬱症之間的關係，至今仍未有一致的結論。這些分歧的研究證據，可能來自於過往研究中有限的樣本數，或是使用單變項分析。本研究嘗試在憂鬱症患者與無相關病史的健康受試者身上，探討其腦部活動與憂鬱程度的關聯。我們從英國生物樣本庫（UK Biobank）納入了417位憂鬱症患者與860位健康受試者。每位受試者都在功能性磁振造影掃描時進行了Hariri 等人（2002）設計的情緒性作業。我們也進一步套用了Schaefer等人（2018）提出的大腦區域-全域分割法執行腦影像的分析。我們觀察到，只有在被情緒性作業激起反應的腦區中，可以觀察到憂鬱程度與腦部活動顯著相關 （在PHQ-9與RDS-4的結果均為 rs = -0.183; p = 0.020）。在單變項模型中，我們透過Wilcoxon等級檢定發現功能性磁振造影訊號對憂鬱症患者與健康受試者，在其憂鬱程度的預測力存在顯著差異。在對PHQ-9的預測上，腹側注意力與感覺運動聯合皮質區網路對憂鬱症患者預測度較好（p < 0.001）、邊緣與預設模式網路則對健康受試者較佳（p < 0.001）。在對RDS-4的預測上，除了額葉頂葉注意力網路外，所有網路都對憂鬱症患者有著較好的預測能力。在對PRS的預測上，則是腹側注意力網路對健康受試者有著較好的預測能力（p = 0.043）。最後，在多變項模型中，我們的結果顯示了效果量與預測度彼此相關（PHQ-9：r = 0.085，p = 0.004；RDS-4：r = 0.062，p < 0.001）。總結來說，本研究的結果為腦部活動與憂鬱程度的預測提供了進一步的證據以及不同的觀點。

In recent years, findings of neural basis for depression have been discovered in both resting-state and task-based functional magnetic resonance imaging (tfMRI). However, the results were largely inconclusive. The disparity may be derived from the limited sample size and the insufficient predictive power using univariate analysis. In the present study, we explored the relationship between brain activity and depressive levels (e.g. current depressive scores, the genetic liability of depression using polygenic risk score), among severe depressive patients (MDD) and healthy controls (HC). We used the tfMRI database from the UK Biobank. Participants (N=417 for MDD; N=860 for HC) performed an emotional task using the Hariri et al., (2002) paradigm in which they viewed faces with negative emotions and shapes. We conducted a parcel-based analysis by employing a local-global parcellation scheme in Schaefer et al., (2018). The correlations between depressive level and brain activity in the task-activated parcels demonstrated heterogeneous responses to emotionally negative stimuli in the cortical network (rs = -0.183; p = 0.020 for the Patient Health Questionnaire 9-question version (PHQ-9) and the recent depressive symptoms (RDS-4) scores in the task-activated parcels). The MDD and the HC groups had different predictive power for depressive levels based on the results of the univariate model. For the PHQ-9, predictive power was significantly larger in MDD than HC groups (p < 0.001) in the ventral attention and motor networks, and smaller in MDD than HC groups (p < 0.001) in limbic and default mode networks. For the RDS-4, predictive power was significantly larger in MDD than HC groups in all seven but frontoparietal control networks. For the polygenic risk score of depression, predictive power in the ventral attention network was larger in the HC than MDD group (p = 0.043). In general, the multivariate model demonstrated better predictive power to depressive level than the univariate model the association between effect size and predictive power . These findings provided evidence for the prediction ability using tfMRI data for depressive levels.