Joint Estimation of Hemodynamic Response Function and Voxel Activation in Functional MRI Data
This paper proposes a method of voxel-wise hemodynamic response function (HRF) estimation using sparsity and smoothing constraints on the HRF. The slow varying baseline drift at the voxel time-series is initially estimated via empirical mode decomposition
- PDF / 1,449,693 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 0 Downloads / 155 Views
Department of Electronics and Communication Engineering, IIIT-Delhi, India Department of Neuroradiology, Neurosciences Centre, AIIMS, Delhi, India {priyaa,anubha}@iiitd.ac.in, [email protected]
2
Abstract. This paper proposes a method of voxel-wise hemodynamic response function (HRF) estimation using sparsity and smoothing constraints on the HRF. The slow varying baseline drift at the voxel timeseries is initially estimated via empirical mode decomposition (EMD). This estimation is refined by two-stage optimization that estimates HRF and slow-varying noise iteratively. In addition, this paper proposes a novel method of finding voxel activation via projection of voxel timeseries on signal subspace constructed using the prior estimates of HRF. The performance of the proposed method is demonstrated on both synthetic and real fMRI data. Keywords: functional Activation detection.
1
MRI,
Hemodynamic
Response
Function,
Introduction
Blood oxygen-level dependent (BOLD) functional magnetic resonance imaging (fMRI) is a non-invasive method to analyze human brain activity under different tasks such as visual, hearing, cognitive, etc [1]. It relates the neural activity with the temporal impulse response which is known as hemodynamic response function (HRF). In this manner, HRF is a proxy measure of underlying neuronal activity in brain. HRF not only varies across multiple subjects, but also in different regions of the brain of a single subject. HRF estimation can play a crucial role in estimating brain voxels activity accurately. In the literature, HRF estimation has been done by two approaches: via regionbased approach and via voxel-based approach [2]. In the region-based approach, regions of interest (ROIs) are extracted by either assuming equally sized regions [3] or via parcellation algorithm [4]. It is assumed that the HRF is same in all the voxels of a region. Hence, the mean of fMRI signal in the ROI is considered for HRF estimation. But, in actual scenario, some voxels may have different HRF within that ROI. Thus, estimated HRF may be suboptimal. In order to overcome the above shortcoming, various methods of voxel-based HRF estimation have been proposed in the literature [5-6]. Here, HRF is assumed to vary across different voxels. However, due to poor signal-to-noise ratio of fMRI time series, c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part I, LNCS 9349, pp. 142–149, 2015. DOI: 10.1007/978-3-319-24553-9_18
Estimating Hemodynamic Response Function and Voxel Activation in fMRI
143
HRF estimates may be potentially misleading. Smoothing in the pre-processing stage can overcome this problem. In addition to the above classification, there are parametric and nonparametric methods of HRF estimation. In the parametric methods, shape of HRF is assumed to be known apriori. However, single nonlinear function is not accurate to model HRFs across the entire brain. Nonparametric methods of HRF estimation do not restrict the shape of HRF and estimate HRF at each time point [5
Data Loading...
