Elsevier

Computerized Medical Imaging and Graphics

Volume 25, Issue 6, November–December 2001, Pages 449-457
Computerized Medical Imaging and Graphics

Lipsia—a new software system for the evaluation of functional magnetic resonance images of the human brain

https://doi.org/10.1016/S0895-6111(01)00008-8Get rights and content

Abstract

This paper describes the non-commercial software system Lipsia that was developed for the processing of functional magnetic resonance images (fMRI) of the human brain. The analysis of fMRI data comprises various aspects including filtering, spatial transformation, statistical evaluation as well as segmentation and visualization. In Lipsia, particular emphasis was placed on the development of new visualization and segmentation techniques that support visualizations of individual brain anatomy so that experts can assess the exact location of activation patterns in individual brains. As the amount of data that must be handled is enormous, another important aspect in the development Lipsia was the efficiency of the software implementation. Well established statistical techniques were used whenever possible.

Introduction

Functional magnetic resonance imaging (fMRI) allows digital images that display local changes in blood flow with a spatial resolution of about 3 mm and a temporal resolution of less than 1 s to be created. Since its invention a few years ago [1], it has become one of the most important technologies used in human brain mapping research.

The data produced by a typical fMRI experiment consist of a time sequence of digital images taken every n seconds, with n=2 being a typical choice. Images consist of several 2D image slices. Usually, two or more experimental conditions are alternated within the same experiment. For instance, some baseline condition may be contrasted with a condition in which some visual or auditory stimulus is presented.

The analysis of fMRI data comprises several aspects. The most important objective is to obtain a statistical parametric map that depicts areas within a subject's brain that show a statistically significant response to an experimental stimulus. In order to obtain such a map, some preprocessing steps as well as elaborate statistical evaluations are needed.

In addition to producing activation maps for individual subjects, methods of inter-subject comparisons are needed so that group studies involving several subjects become possible. This, in turn, requires methods of spatial normalization and co-registration where data acquired from different subjects are geometrically aligned. Lastly, the results obtained from the statistical analysis must be visualized graphically so that they can be presented to the scientific community.

This article describes the various aspects of fMRI data analysis as implemented in a new non-commercial software package called Lipsia (Leipzig Image Processing and Statistical Inference Algorithms). The Lipsia software was designed to enable the routine processing of large numbers of large data sets. Lipsia currently functions in an environment in which fMRI experiments are conducted for about 10 h daily generating about 2 Gigabyte of data every day. As new imaging hardware that allows greater spatial resolution becomes available, this number may increase by a factor of two in the near future.

Lipsia was designed with the following objectives in mind. Clearly, it must be able to handle such large amounts of data efficiently. Furthermore, as the data processing is performed by scientific personnel such as psychologists or medical doctors who are experts in neither computer science nor image processing, the software must be easy to use and robust. Another very important objective was to support visualizations of individual brain anatomy so that experts can assess the exact location of activation patterns in individual brains.

This article is organized as follows. We begin by describing the various types of data pertaining to an fMRI experiment. We then explain the processing chain that these data are subjected to. Algorithms that are described elsewhere in the literature are briefly summarized. New algorithms are described in more detail.

Section snippets

Data types

In the course of an fMRI experiment, three or sometimes four different types of data are generated.

The first data type are the functional data consisting of a time series of scans where each scan consists of a stack of 2D parallel image slices. The most common imaging technique is called Echo Planar Imaging (EPI). In EPI, only one slice can be acquired at a time. Typical imaging parameters are an acquisition time of 125 ms per slice, with 16 slices constituting one image scan. The image matrix

Preprocessing

As noted above, fMRI slices are usually not acquired simultaneously. To correct for the temporal offset between the slices acquired in one scan, Lipsia offers both a linear interpolation and a sinc-interpolation algorithm. The sinc-interpolation algorithm is based on the Nyquist–Shannon Theorem.

Baseline drifts are eliminated by a high-pass filter. The cutoff frequency is specified by the user. As an alternative, Lipsia also offers a high-pass filter as part of the statistical modeling.

Spatial transformations

In order

Software performance

The Lipsia software is implemented in C and C++ and runs within the Unix operating system. Programs are started either from the command line, from a Unix-shell-script or from a mouse-driven graphical user interface. Most programs additionally run in a parallelized version implemented on a 16-processor shared-memory parallel computer (SGI origin 2000).

In particular, the preprocessing (slicetime and baseline correction) as well as volume rendering and the statistical analysis run in parallel. As

Discussion

We have introduced the software package Lipsia that is used to process fMRI data. In view of the large amounts of fMRI data, one of the most important aspects in developing Lipsia was computational efficiency and ease of data handling. This was achieved by using a versatile data format and efficient and, in part, also parallel implementations.

Various other software packages for the analysis of fMRI data exist, e.g. [25], [26], [27], [28]. The most widely accepted statistical methodology for

Dr Gabriele Lohmann received her diploma in mathematics and mathematical logic from the University of Münster in 1984, her doctorate in computer science from the Technical University of Munich in 1991, and her habilitation in applied computer science in 1999. She spent an academic year at Indiana University, Bloomington, IN, supported by a Fulbright scholarship, and a six-month research stay at the Computer Vision Lab of the University of Massachusetts, Amherst, MA. From 1984 until 1991, she

References (29)

  • P. Talairach et al.

    A stereotactic coplanar atlas of the human brain

    (1988)
  • Kruggel F, Lohmann G. Automatical adaptation of the stereotactical coordinate system in brain MRI data sets. In: Duncan...
  • W.H. Press et al.

    Numerical recipes in C

    (1992)
  • O. Josephs et al.

    Event-related fMRI

    Human Brain Mapping

    (1995)
  • Cited by (323)

    View all citing articles on Scopus

    Dr Gabriele Lohmann received her diploma in mathematics and mathematical logic from the University of Münster in 1984, her doctorate in computer science from the Technical University of Munich in 1991, and her habilitation in applied computer science in 1999. She spent an academic year at Indiana University, Bloomington, IN, supported by a Fulbright scholarship, and a six-month research stay at the Computer Vision Lab of the University of Massachusetts, Amherst, MA. From 1984 until 1991, she was a researcher at the German Aerospace Research Center working in the field of satellite remote sensing. She is currently a scientist at the Max-Planck-Institute of Cognitive Neuroscience where she leads a research group that specializes in mathematical methods of fMRI data analysis. Her research interests include computer vision, pattern recognition and neuroscience.

    Dr Karsten Müller received his PhD in mathematics from the Leipzig University in 1996. He spent a year working in the field of interpolation theory at the Department of Computer Science of the Leuven University. During that time he got the promotion prize of the German Academy of Sciences Leopoldina. Supported by the German Research Society, Karsten Müller worked two years at the Faculty of Mathematics at the Chemnitz University of Technology and published several papers on special subclasses of matrix-valued meromorphic functions. He is currently working on mathematical methods in fMRI at the Max-Planck-Institute of Cognitive Neuroscience, Leipzig.

    Dr Volker Bosch received his MS degree in psychology from the Free University, Berlin, Germany, in 1995. In 1999, he received his PhD degree from the University of Leipzig. He currently holds a scientific staff position at the Max-Planck-Institute of Cognitive Neuroscience. His research interests include the statistical methods used in fMRI and EEG.

    Heiko Mentzel is a student at the computer science department of University of Leipzig. Since 1999, he has also been a student assistant at the Max-Planck-Institute of Cognitive Neuroscience, Leipzig.

    Sven Hessler finished his study of information technology at the University of Cooperative Education, Glauchau, in 1999. Afterwards, he worked as a software developer at the Max-Planck-Institute of Cognitive Neuroscience. He is currently working for a software firm in London.

    Lin Chen received his master's degree in mathematics from Beijing Institute of Applied Physics and Computational Mathematics, China, in 1994. He received his diploma in industrial mathematics from the University of Kaiserslautern, Germany, in 1998. He is a PhD candidate at the Max-Planck-Institute for Mathematics in the Sciences, Leipzig, Germany. His current research interest covers the theory of neural networks and their applications to 3D image understanding and computer vision.

    Professor Dr Yves von Cramon studied medicine in Munich and Giessen. After receiving his MD he joined the Max-Planck-Institute of Psychiatry in Munich were he earned the Postdoctoral Qualification in Neurology in 1979. From 1984 to 1994 he built up the Department of Neuropsychology at the City-Hospital Bogenhausen (Technical University of Munich). During these years (1986) Dr von Cramon was offered an appointment as Professor of Neurology. In 1994, he became Scientific Member and Director (Department of Neurology) at the newly founded Max-Planck-Institute of Cognitive Neuroscience (Leipzig). Since 1996, he has also been Professor of Neurology at the University of Leipzig where he is responsible for the Daycare Clinic for Cognitive Neurology. His main research fields are: functional neuroanatomy; cognitive neurology; neurological rehabilitation; and clinical neuropsychology.

    View full text