Model description¶
Blind Deconvolution Analysis (BDA)¶
HemoLearn is a Python module offering a new algorithm that aims to fit a rich multivariate decomposition of the BOLD data using a semi-blind deconvolution and low-rank sparse decomposition. The model distinguishes two major parts in the BOLD signal: the neurovascular coupling and the neural activity signal.
Mathematically, if we have a single subject with \(P\) fMRI time series of length \(\widetilde{T}\), if we share the spatial maps, the considered data model is:
We aim to distangle the neurovascular coupling modelled by \(\sum_{m=1}^{M} \boldsymbol{\Theta}_m ^\top \boldsymbol{v}_{\delta_{im}}\) from the neural activation signals modelled by \(\sum_{k=1}^{K} \boldsymbol{u_k}^\top \boldsymbol{z_{ik}}\) by minimizing the following cost-function:
With \(\lambda_i\) being the temporal regularization parameter for the i-th subject, \(\eta\) the spatial sparcity parameter, \(K\) the number of neural components and \(M\) the number of vascular regions considered.