A new spatially constrained NMF with application to fMRI

In this paper the problem of BOLD detection is addressed. The focus here is on non-negative matrix factorization (NMF), which is a data driven method and able to provide part-based representation of data. A new constrained optimization problem is proposed for the purpose of BOLD detection. The proposed constraint imposes some prior spatial information of active area inside the brain, on the decomposition process. The constraint is built up based on the type of stimulus and available physiological knowledge of the brain performance. The simulation results on both synthetic and real fMRI data show that applying the proposed constraint improves the BOLD detection performance.

I. Introduction

Functional Magnetic Resonance Imaging (fMRI) is an imaging technique which provides useful information of brain functioning. The main goal of fMRI processing techniques is detecting the blood oxygen level dependent (BOLD) response as the result of brain activity [1]. Activation detection in brain functional imaging experiments helps to retrieve the neuronal activity as much as possible in response to cognitive or behavioral stimuli tasks. In general, fMRI analysis methods are divided into two groups: model-based techniques and data-based techniques. Model-based methods need prior knowledge about the activation task-waveform to work properly. In contrast, data-based methods which also called model-free do not rely on any assumed model or prior knowledge. However, exploiting some available a priori information can improve the performance of these methods.