FSL in MEDx

This new module is a collection of functional and structural brain image analysis tools developed by the Image Analysis Group at Oxford University under the leadership of Dr. Stephen Smith. The structural tools consist of an automatic brain extraction tool (BET), a Linear Registration Tool (FLIRT), and a noise reduction technique based on Smallest Univalue Segment Assimilating Nucleus (SUSAN). The Easy Analysis Tool (FEAT) provides a very straightforward and easy-to-use interface for the fMRI data analysis process. This interface enables the user to specify pre-processing steps such as motion correction, spatial filtering, global intensity normalization, and temporal filtering. The user can also specify a paradigm and the statistical method for individual or group analyses. The functional analyses employ a General Linear Model and a model-free approach for cases where no a priori knowledge of the expected response exists. FSL also includes tools for analyzing event-related fMRI experiments. The FSL module provides a powerful, fast, and robust alternative set of tools that complement and enhance the existing MEDx tool set.

Brain Extraction Tool (BET)

An accurate registration often requires that non-brain tissue such as eyeballs, skin, and fat be removed from MR images prior to registration. Brain/non-brain segmentation is also essential for brain atrophy measurements which quantify brain volume changes with respect to some normalizing volume such as the skull or head size.

Traditionally, the outline of the brain surface is delineated manually or through interactive region-growing algorithms, both of which require the user to specify a seed point and various thresholds. The time constraints, the need for training, and the lack of reproducibility work against interactive methods.

The fully automatic brain extraction algorithm in FSL belongs to a class of sophisticated models which utilize deformable surfaces. The BET technique finds the brain surface by adjusting the vertices of a tessellated sphere. In finding the surface, each vertex is moved through iterative incremental surface updates.

		---> ---> --->
	1A. Rendered brain without deskulling		1B. Rendered brain after BET deskulling

		---> ---> --->
	2A. Pre-extraction coronal slice of the brain		2B. Post-extraction coronal slice of the brain

	3A. Some coronal slices before using BET
	3B. Those same slices after using BET

FEAT - FMRIB's Easy Analysis Tool

A complete fMRI analysis is indeed a "feat" consisting of pre-processing steps, statistical analyses, and comparison of the analyses between different subjects and sessions. FEAT accomplishes all this in a single run based on the specifications in a compact GUI. The FEAT menu provides options for different types of analyses. The FILM model-based option is a sophisticated implementation of the General Linear Model (GLM). This option can be used for block or single-event designs. The design matrix of the experiment can be generated using the Simple model setup or Full model setup options. An alternative analysis method is the IRVA semi-model-free approach which is based on inter-repetition variance analysis (IRVA). IRVA comes in two varieties: the Cyclic IRVA is designed for regularly spaced stimulations, and the Triggered IRVA for irregularly spaced experiments. Below are some screen shots of the FEAT GUI.

	4. The top-level FEAT interface.
	5. Dialog box for creating simple models
	6. General Linear Model setup

FSL is released under the GNU General Public License. The source code for the version used in MEDx (v1.3) is available here.

Top