神経データ解析 Neuronal Data Analysis

P2-1-244 時間空間拘束条件として解剖学的結合の事前知識を取り入れた脳磁図電流源推定法 MEG source reconstruction spatiotemporally constrained by prior knowledge of anatomical connectivity ○福嶋誠1,2,3, 山下宙人2佐藤雅昭2 ○Makoto Fukushima1,2,3, Okito Yamashita2, Thomas R. Knösche4, Masa-aki Sato2 奈良先端大院・情報・計算神経科学1, ATR脳情報解析研2, 日本学術振興会3, マックスプランク認知神経科学研4 CNS Lab, Grad Sch of Info Sci, NAIST, Nara, Japan1, ATR NIA Labs, Kyoto, Japan2, JSPS, Tokyo, Japan3, MPI CBS, Leipzig, Germany4 We propose an MEG source reconstruction method with a spatiotemporal constraint of anatomical connectivity over the whole brain. The source dynamics model (i.e., the constraint on the sources) is based on neuroanatomical connections inferred from diffusion MRI. Previously, the source dynamics model took local and/or self connections into account. The proposed method incorporates long-rage anatomical connections to the source dynamics model as well. The source dynamics is represented by the multivariate autoregressive (MAR) model, in which MAR matrix entries (i.e., functional connectivity between sources) to be estimated are restricted by anatomical connectivity. Namely, MAR entries corresponding to anatomically connected source pairs are parameterized, and the remaining ones are fixed to zero. The model parameters and source amplitudes are jointly estimated using the variational Bayesian method. Through simulation, we compare the accuracy of the proposed method to that of its non-dynamical counterpart. We confirmed the proposed method suppressed spurious sources and, unlike the non-dynamical method, it recovered a low amplitude source at low signal-to-noise ratio. Functional connectivity estimated by the proposed method was also more accurate than that obtained from the non-dynamical method. The real applicability of the proposed method is investigated using a face recognition dataset for one subject. The proposed method estimated cortical activity in line with the literature. Especially, it discovered the activity in one of face selective regions not estimated by its non-dynamical counterpart. The number of functional connections identified by the proposed method was larger than that of the non-dynamical method, and these connections would help to reconstruct the missing activity. In future, we will examine subject commonality and variability of the obtained results. Effects of false anatomical connectivity priors on the estimates should be investigated systematically.

上部に戻る

前に戻る