MELODIC - User Guide

(Multivariate Exploratory Linear Optimized Decomposition into Independent Components)
ICA-Based Model-Free Analysis of 4D Data
MELODIC Version 1.0

INTRODUCTION

MELODIC (Multivariate Exploratory Linear Optimized Decomposition into Independent Components) uses ICA (Independent Component Analysis) to decompose 4D data (for example FMRI data) into different interesting spatial and temporal components. It can pick out different activation and artefactual components without any model being specified. Thus the output is a new 4D data set (each volume being a separate spatial component) and a text file of separate time courses (temporal components).

For more detail on MELODIC and an updated journal reference, see the MELODIC research web page. If you use MELODIC in your research, please quote the journal reference listed there.

The different MELODIC programs are:


PREPROCESSING

Before running MELODIC, you probably want to have carried out some preprocessing on this data, principally to correct for head motion and to remove slow drift in the data. This is easy using FEAT. From the FEAT GUI, carry out the following:


Melodic GUI

MAIN OPTIONS

Select the 4D input data which should be a 4D Analyze format image.

After selecting the input data, the GUI will tell you how many volumes (time points) are in your data. You need to choose how many Output components MELODIC will create. If you choose too many, ICA will take a long time to run (it may even fail to finish) and also interesting components (such as single activation components) may be "incorrectly" split across several components. If you choose too few, then ideally separable components will get merged. The default is to use 1/4 of the number of input time points. In the future we will include an automatic dimensionality estimation tool to increase the sophistication of this choice.

When you are ready press GO. MELODIC will put all results in a new directory, whose name is derived from the input data filename.

ADVANCED OPTIONS

By default BET brain extraction is used to mask out non-brain parts of the image before ICA is run. You can optionally turn this off by pressing the button next to Use BET to mask data. If you do this, you instead need to select a Threshold % with which to threshold the data (although this can be set to 0 if you really want to process all of the data).

By default MELODIC will Variance normalise timecourses. This means that every voxel's timecourse will be rescaled to have unit variance, in order to allow ICA to find interesting components, even if they do not have high temporal variance. This preprocessing step can be turned off if desired.

If you do not wish the web page report to be created, press the button next to Create web page report.


melodic COMMAND-LINE PROGRAM

Type melodic --help to get usage.


melodicreport COMMAND-LINE SCRIPT

Type melodicreport to get usage. This script processes the melodic output data to create the web page reports.


Copyright © 2001, University of Oxford. Written by C. Beckmann and S. Smith.