Application Provides Functional Enhancement for fMRI

Computer engineers have now developed a more robust, three-stage approach to fMRI that could significantly improve the detection of neural activity and allow researchers to make more precise interpretations of fMRI data.

A new way to analyze fMRI data, which could improve was reported in 2010 (no. 3) issue of the International Journal of Computational Biology and Drug Design. Scientists have long known that changes in blood flow and blood oxygenation in the brain (hemodynamics) are correlated with activity in brain cells, neurons. When a neuron is active, it demands more energy from glucose and this need increases blood flow to the regions of the brain where there is more neural activity. This leads to local changes in the relative concentration of oxyhemoglobin and deoxyhemoglobin and alterations in local cerebral blood volume and in local cerebral blood flow, which researchers have been measuring using fMRI since the early 1990s. Since then, brain-mapping using this comparatively noninvasive technique, which also avoids exposure to ionizing radiation has become more widely used.

Researchers have used fMRI to examine brain development and function, to diagnose problems following injury and to predict when a person might be healthy enough to return to work, as an alternative to lie detectors, to peer, allegedly, into an individual's dreams, and even to communicate with patients in a vegetative state. Many of the experiments that have received attention are controversial in that interpreting images of altering blood flow in the brain is only a proxy of actual activity Moreover, extrapolating those proxy images to thoughts and behavior involves a not insignificant extrapolation.

Dr. Chuan Li and Dr. Qi Hao, from the department of electrical and computer engineering at the University of Alabama (Tuscaloosa, USA), developed the new technique. The team explained that there are three stages to their approach: prediction, modeling, and inference. Prediction involves identifying regions of interest associated with an extraordinary amount of neural activity through temporal clustering analysis (TCA). Modeling involves categorizing the fMRI signals related to neural activity into event prototypes through linear predictive coding (LPC). Finally, inference is the determination of the types of neural activity taking place in terms of activation, deactivation, and normality using a type of statistical analysis known as Bayesian inference.

The approach bypasses to some extent the problems inherent in current approaches to fMRI, namely low signal-to-noise ratio, high data volumes, differences between patients or subjects, and artifacts caused by the movement of the person being scanned. Their application allows them to turn large amounts of often-noisy data into discrete sequences of neural activity events. The investigators have shown how well their approach works by analyzing data from fMRI scans on volunteers involved in the simple activities of drinking a glass of water or a glass of glucose solution. "Our expertise is in signal processing and machine learning. Our research goal is to develop a set of powerful signal processing tools for fMRI researchers,” concluded Dr. Hao.

Related Links:
University of Alabama