Analysis of functional MRI data

Raw functional MRI data must undergo several critical processing steps before yielding the (indirect) brain activation images:

• Preprocessing: the images are smoothed to reduce noise and the artifacts (movements, orientation and spatial distortion) are corrected.
• Normalization: this is necessary to compare patient examinations that are different or performed at different times. The images are repositioned either between two examinations, or to match a reference atlas (Talairach), to overlap in the same spatial coordinate.
• Statistical analysis: this is based on mathematical modeling of the expected hemodynamic response, which will depend on the paradigm used. The most common type of model is the generalized linear model (GLM). This model will examine each pixel to detect those whose signal variation in time is linked to the sequence of different activation tasks. The pixels considered as statistically significant can then be represented overlapping high-resolution morphological imaging for better location.