An automatic tool to facilitate the statistical group analysis of DTI
Introduction
Neuroimaging has been increasingly applied to patients with brain disorders in order to investigate in vivo abnormalities in the brain [1], [2], [3], [4]. The analysis and visualization of medical image data extracted from neuroimaging is crucial for prevention and diagnosis. Recent neuroimaging techniques such as Diffusion Tensor Imaging (DTI) have arisen in order to characterize the structure of the brain. Fractional Anisotropy (FA), a measurement extracted from DTI, describes the fibre structure (Fig. 1) and is used to analyse disease abnormalities in the white matter and compare FA between subjects in clinical research.
In clinical studies, after preprocessing the images, a statistical analysis between groups is usually performed to analyse the differences between subjects׳ brains due to the abnormalities caused by disease or aging. There are several techniques that can be used to carry out the analysis [5], [6], including region of interest (ROI) analysis, tract analysis and voxelwise analysis.
Earlier studies have mainly used the ROI method [7], [8], [9]. BrainSuite [10], Brain Voyager QX [11] and FreeSurfer [12] include this technique. However, the fact that in this method the regions are drawn by hand makes it difficult to define ROIs between subjects in a reliable and accurate way [13]. Additionally, this method can miss significant information in non-selected regions.
More sophisticated methods use tractography, which is the tracking of the brain fibres, in order to compare FA values across subjects. In this approach, called tract analysis, white matter fibres are traced using ROIs. The UNC-Utah NA-MIC DTI framework [14] is an example of a tract-based tool based on Goodlett׳s quantitative tractography method [15]. The accuracy of the methods depends on the registration of each subject to the standard space. In addition, only the information that is extracted from the tracts is analysed.
Voxelwise analysis (VBA) methods attempt to solve the various limitations of the other methods by analysing the whole brain, which means that it is not necessary to specify any regions by hand. Moreover, automated registration is carried out to match all the images. Due to these improvements, several recent studies have used voxelwise analysis. Statistical Parametric Mapping (SPM) [16] and Tract-based Spatial Statistics [17], the latter being part of FSL Software [18], [19], are two examples of voxelwise analysis software. Both methods differ in the spatial normalization of FA images and in the parameters applied in the smoothing. TBSS was developed to alleviate the alignment and smoothing problems of SPM. Despite some weaknesses in revealing tract specific features, TBSS has become the most popular tool for analysing DTI images.
Recently, several pipelines have been attempting to facilitate the processing of DTI datasets. Some of them provide an environment in which to design workflows within and between packages, and users need to construct pipelines that contain the steps and dependencies (MIPAV [20], JIST [21], Nipype [22] and LONI [23]). These types of powerful tools are designed for expert users. However, most of the time the target users are novices in programming and they may have difficulty processing DTI datasets and calculating group statistics. The recently published PANDA [24] does not require that the pipeline be defined, and it aims to overcome the handicap of complexity by adding a GUI. However, a usability study with end users is missing to objectively evaluate the tool׳s real ease of use. Usability testing is essential for determining whether a solution works with target users [25], [26], [27], [28]. Moreover, the authors of PANDA do not detail their design methodology. Following a design methodology, such as the user-centred methodology, will provide a full understanding of the problems in existing tools and facilitate the development of a tool that fulfils the needs of the users.
Despite the appearance of pipelines, TBSS, or FSL in general, is the preferred tool amongst clinicians for statistical group analysis [29], [30], [31], [32]. However, FSL requires advanced knowledge of DTI analysis, and due to its numerous features users are required to master the software. Therefore, they often get confused, as each step must be done manually. In addition, DTI statistical group analysis normally takes hours and sometimes more than one tool is needed for a complete analysis, further increasing the complexity of clinical research studies. For all these reasons, it is very important to automate the process in order to reduce user intervention and consequently minimise the number of errors, analysis time and costs.
In this paper, we present a novel medical tool, which we call DTIStatistics, for automatic statistical analysis based on FSL, MRICron [33] and R [34]. The main objective of DTIStatistics is to enable easy preprocessing and the calculation of group statistics from DTI datasets. Our proposal has a friendly graphical user interface and its automated pipeline minimises the need for user intervention. This tool was designed using a user-centred methodology [35]. Following this methodology, we performed an iterative usability evaluation with expert users in Human–Computer Interaction, Bioengineering and Medicine in order to create a suitable tool. Once the software met the experts׳ requirements, we validated the final version of the tool with our target users. To that end, we recruited Biomedical Engineering students to validate the tool, and the validation confirmed that our tool facilitates the preprocessing and statistical group analysis of DTI datasets. In order to demonstrate the validity of DTIStatistics in clinical research, we present example results obtained with our tool in the study of late-onset major depressive disorder in the elderly.
Section snippets
Design
In order to create a suitable tool for the final user, we followed a user-centred design methodology, which has been successfully applied in other recent studies [36], [37], [38]. This methodology tests which aspects of the system are adequate in order to improve the design. This methodology not only requires that designers analyse how the product is going to be, but that they also test their initial proposals in a realistic validation process with representative users. For this purpose we used
Validation results
The validation that was carried out with Biomedical Engineering students was divided into two parts: an experimental validation comparing a typical pipeline and our tool, and a questionnaire related to measuring the satisfaction of the target users.
In the experimental validation, all the users finished the three tasks using DTIStatistics (see Table 2). In contrast, when using FSL, all the users finished the first task and 28 of 30 finished the second one; no user was able to complete the third
Discussion
This study presents the design and evaluation of a novel tool called DTIStatistics for the automatic statistical group analysis for DTI. Our results confirmed that our novel tool overcomes the problems defined by experts using existing systems and provides a fully automated and significant easy-to-use pipeline for the preprocessing and statistical analysis of DTI images using FSL, MRICron and R. The choice of these software tools was motivated due to the fact that they have become a reference
Conclusion
This article describes the development of a software tool as the solution to a specific problem in clinical research, using formal methodologies to design and evaluate such a tool. This research also demonstrates that the statistical analysis of DTI images using software of wider application like FSL can be significantly sped up by using an ad hoc tool. Developing a more usable and efficient software tool is worthwhile as it leads to shorter analysis times, which means lower costs. Furthermore,
Summary
Analysing and visualizing DTI images is essential in the prevention of neurological diseases. Various powerful tools have been established for this task; however, due to the complexity of the existing tools, users may have difficulty analysing DTI images and calculating group statistics. Users need to master several tools, and user intervention is required in every step involved. Automating the pipeline could reduce analysis times and possible errors. Therefore, an automatic and easy-to-use
Acknowledgements
The authors would like to thank the “Beca Universidad de Navarra” from the University of Navarra, the University Hospital, University of Navarra, and QPEA Association for funding this work. We want also to thank to the expert committee and the Biomedical Engineering student volunteers that helped us to validate the tool. Additionally, we want to show gratitude for the developers of FSL, MRICron and R.
References (51)
Neuroimaging trends and future outlook
Neurol. Clin.
(2014)- et al.
Structural neuroimaging in aging and Alzheimer׳s disease
Neuroimaging Clin. N. Am.
(2012) - et al.
Neuroimaging of dementia
Neurol. Clin.
(2014) - et al.
A review of diffusion tensor magnetic resonance imaging computational methods and software tools
Comput. Biol. Med.
(2011) - et al.
Diffusion tensor imaging (DTI) and proton magnetic resonance spectroscopy (1H MRS) in schizophrenic subjects and normal controls
Psychiatry Res.
(2001) - et al.
BrainSuite: an automated cortical surface identification tool
Med. Image Anal.
(2002) BrainVoyager — Past, present, future
Neuroimage
(2012)FreeSurfer
Neuroimage
(2012)- et al.
Group analysis of DTI fiber tract statistics with application to neurodevelopment
Neuroimage
(2009) - et al.
Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data
Neuroimage
(2006)