Missing data estimation in fMRI dynamic causal modeling

Abstract

Dynamic Causal Modeling (DCM) can be used to quantify cognitive function in individuals as effective connectivity. However, ambiguity among subjects in the number and location of discernible active regions prevents all candidate models from being compared in all subjects, precluding the use of DCM as an individual cognitive phenotyping tool. This paper proposes a solution to this problem by treating missing regions in the first-level analysis as missing data, and performing estimation of the time course associated with any missing region using one of four candidate methods: zero-filling, average-filling, noise-filling using a fixed stochastic process, or one estimated using expectation-maximization. The effect of this estimation scheme was analyzed by treating it as a preprocessing step to DCM and observing the resulting effects on model evidence. Simulation studies show that estimation using expectation-maximization yields the highest classification accuracy using a simple loss function and highest model evidence, relative to other methods. This result held for various dataset sizes and varying numbers of model choice. In real data, application to Go/No-Go and Simon tasks allowed computation of signals from the missing nodes and the consequent computation of model evidence in all subjects compared to 62 and 48 percent respectively if no preprocessing was performed. These results demonstrate the face validity of the preprocessing scheme and open the possibility of using single-subject DCM as an individual cognitive phenotyping tool.

From correlation to causation: Estimating effective connectivity from zero-lag covariances of brain signals

“…The threshold is derived from the 10% strongest connections, disregarding their signs. As there is generally no full ground truth for the connectivity inferred from human fMRI recordings available [35,31,32], we cannot definitely assess the degree to which the result of our inference are correct. We can, however, establish whether they are plausible.…”

The influence of EM estimation of missing nodes in DCM on model ranking

“…This paper extends our previous paper [9] by demonstrating the effect of the preprocessing Expectation Maximization (EM) estimation of missing nodes scheme on model ranking and averaging.…”

“…The EM algorithm that was used for the estimation of missing nodes is described in detail in [9]. Given the available data X, the missing data Z, and the unknown parameters θ, along with the likelihood function L(θ; X, Z) = p(X, Z|θ), the maximum likelihood estimate of the unknown parameters is determined by the marginal likelihood of the available data.…”

“…In the occurrence of a missing node, the EM estimation scheme was applied to obtain a time series. Once the inputs (via the design matrix) and the time series were defined, DCMs were specified and estimated according to the 16 models described in our previous publication [9] for the Go/NoGo and Simon tasks respectively. Alternative models were specified, estimated, and compared against all other models for each subject for both datasets.…”

Characterizing the theory of mind network in schizophrenia reveals a sparser network structure

“…Accordingly, recent research suggests that mentalizing can be considered a multidimensional construct, subsuming and integrating different sets of cognitive mechanisms [22,77,78], which might be particularly relevant for patients with schizophrenia [43]. Along this line, it should be considered that in total seven participants failed to engage specific nodes within the specified network, which is a common issue in DCM studies [79]. Although several reasons can account for an individual's missing activation of a specific brain region [80], it might be that the excluded participants used an alternative strategy to perform the task and thus recruit different brain regions.…”