Sparse network-based models for patient classification using fMRI

Abstract

Pattern recognition applied to whole-brain neuroimaging data, such as functional Magnetic Resonance Imaging (fMRI), has proved successful at discriminating psychiatric patients from healthy participants. However, predictive patterns obtained from whole-brain voxel-based features are difficult to interpret in terms of the underlying neurobiology. Many psychiatric disorders, such as depression and schizophrenia, are thought to be brain connectivity disorders. Therefore, pattern recognition based on network models might provide deeper insights and potentially more powerful predictions than whole-brain voxel-based approaches. Here, we build a novel sparse network-based discriminative modeling framework, based on Gaussian graphical models and L1-norm regularized linear Support Vector Machines (SVM). In addition, the proposed framework is optimized in terms of both predictive power and reproducibility/stability of the patterns. Our approach aims to provide better pattern interpretation than voxel-based whole-brain approaches by yielding stable brain connectivity patterns that underlie discriminative changes in brain function between the groups. We illustrate our technique by classifying patients with major depressive disorder (MDD) and healthy participants, in two (event- and block-related) fMRI datasets acquired while participants performed a gender discrimination and emotional task, respectively, during the visualization of emotional valent faces.

A Review on Dependence Measures in Exploring Brain Networks from fMRI Data

“…Therefore, the functional connectivity has many beneficial usages for medical purposes. For instance, the use of functional connectivity to discriminate between schizophrenia patients and healthy persons in [23] or between Alzheimer patients and healthy persons in [24]. Exploring functional connectivity is roughly divided into two approaches depending on the method of defining brain regions i.e., seed method and model-free method, which are previously mentioned in section 2.…”

“…If x i and x j are partially uncorrelated i.e., (Σ −1 ) i j = 0, x i and x j are also conditionally independent with given x k . The work in this approach therefore estimates the covariance matrix from the fMRI data and reveal the zero pattern in the inverse of covariance matrix [35,23,25].…”

“…In other words, the corresponding estimation problem is the minimization of distance function (10) subject to constraints in (9) and the zero constraints in A. The work that uses conditional independence as a measure for exploring the functional connectivity among regions by observing zero pattern of the inverse covariance matrix is found in [93,23,25,35,94]. They used Gaussian log-likelihood function with ℓ 1 -regularization as objective function to obtain a sparse inverse covariance matrix.…”

“…Table 1 and 2 provide examples of papers that applied each of the dependence measures for infering the connectivity (some paper may not experiment on the fMRI application) where we categorize the inference methods as explained in section 5. [30,37,24] conditional independence statistical test [95] constrained parametric estimation [23,25,35] path matrix of SEM statistical test [48,47,20,66] constrained parametric estimation [48,47,20,66] …”

Comparing functional connectivity based predictive models across datasets

“…For this, analysis pipelines are used to estimate the connectome from rs-fMRI, and then to build a classification model that predicts the clinical group from the resulting FC estimates. Such a pipeline has been introduced in [5], where sparse FC network estimates are used to classify subjects with major depression disorder from healthy subjects. A fully-automated pipeline has been proposed recently to highlight autism biomarkers [6] from large and multi-site autism datasets.…”