Cerebral Cortical Thickness Estimation using the TINA Open-Source Image Analysis Environment
Abstract
This paper gives an overview of the use and development of the TINA open-source medical image analysis environment, with respect to the determination of human cerebral cortical thickness estimation from magnetic resonance images. The ultimate aim of TINA is to provide a validated system where the source code and datasets are freely available in order to allow peer-validation of published results.
Keywords
Source Code and Data
No source code files available for this publication.
Reviews
Stephen Aylward
Sunday 18 September 2005
Summary: An open-source, view-based (i.e., 2D) image processing system. Ultimate focus is medical image analysis, but also includes routines useful to computer vision (camera geometry, stereo matching).
Written in C with some platform specific code for UNIX/Linux/Cygwin.
The system was largely developed within one group and is perhaps limited in functionality. It is not appropriate for 3D medical image analysis application development and contains a limited number of methods for 2D image analysis.
Most of the information below was gathered from the TINA website, not from the paper. The TINA website is excellent, but it revealed the limitations of the system that weren’t specifically mentioned in the paper. The TINA website is http://tina-vision.net The website appears very up-to-date - with many papers from 2005. Perhaps, however, the paper is discussing a verision of TINA that supports 3D image analysis and more functions. If so, I apologize for my mistake, and I look forward to clarification from the authors.
Hypothesis: The authors propose that TINA is an effective tool for medical image analysis. In particular, it is useful for cortical thickness estimation.
Evidence: Exploring the TINA website suggests that TINA has limited capabilities for representing images and for providing methods that work on multiple types of images. Most notable, (1) a TINA image is a C record that uses a (void *) variable to support different pixel types, (2) it only supports 2D images, and (3) special functions must be written to access vector pixel types.
Regarding, image processing, TINA appears to only provide a small set of methods. Consider, for example, that the following “Noise Filters” are currently available (the following is the documentation from their website):
The following routines can be used for the removal of various forms of noise and unwanted data variation from images.Imrect *im_bthresh(double k,Imrect *im)
Returns an image which is a binary thresholded version of the input image around the value defined by 6#6.
Imrect *im_thresh(double k,Imrect *im)
Returns an image which has been thresholded to set all pixel values below 6#6 to zero. This is useful, for example, in the elimination of unwanted terms in the 2D Fourier domain of an image.
Imrect *im_corrupt(Imrect * im, int dx, int dy, double a, double b)
Generate an image by adding uniform random noise (generated by imf_unif_noise(im->width, im->height, dx, dy, a, b)) to the input image. Useful in the generation of simulated test data for the evaluation of algorithm stability.
Imrect *im_rank(Imrect *im, int range, double noise)
Returns an image with each pixel (of measurement accuracy 11#11 ) given by the rank of its value in the surrounding 12#12 x 12#12 patch. Useful for enhancing the spatial information content of an image before the process of cross correlation or template matching.
Imrect *im_median(Imrect *im)
Returns a median filtered (discontinuity preserving) version of the input image (each pixel replaced by the median value of the 3x3 neighbourhood). Useful for the removal of sensor or aliasing derived pixel dropout. This technique should only be used when there is no alternative, generally it is better to eliminate the problem at source).
Imrect *im_tsmooth(Imrect * im1)
Returns an image which has been smoothed by averaging in a direction tangential to the image gradient at each point (discontinuity preserving).
Regarding the specific application for cortical thickness estimation, it is interesting that their volume estimation analysis must have been conducted by processing the slices individually and not as a 3D segmentation task. It would be interesting to compare those estimates with estimates made when 3D smoothing and other 3D neighborhood information is considered.
In summary, their work is extremely well presented in this paper and in the many other papers presented on their website. Their work is also extremely important and beneficial - clinically and to the image analysis community. This software package, however, is perhaps a bit limited in scope.
Open Science: The authors and the TINA project have made use of and contributed to open-source. Their system integrates with R and promotes other open-source packages (gnuplot, etc.). They are also making many of their datasets available online.
However, their software product, TINA, is not the most up-to-date image analysis software package available. Users should consider VXL, ITK, MITK, MeVisLab or other pacakges that are more easily extended to other medical image analysis tasks.
Sylvain Bouix
Thursday 8 September 2005
Summary: This article presents a very interesting toolkit used for cortical grey thickness measurement. The paper gives a very thorough description of the the open source software, TINA, but gives almost no information on the thickness measurement itself. After reading the referenced technical report, thickness is measured by shooting rays normal to the GM/WM boundary and carefully measuring their lengths.
Hypothesis: Thickness is measured as the length of a straight line connecting the GM/WM surface to the GM/CSF surface.
Evidence: The method is validated as follows. The authors made thickness measurement in 13 young adults and compared their results to the ones of Kabanis et al. made on 20 young adults on a different data set. This form of validation is questionable, but due to the lack of a true gold standard for cortical thickenss measurements it is already quite an achievement that the method is validated.
Open Science and Reproducibility: Source code is provided as well as user’s and developper’s guides. The data is not shared. I have not tried to reproduce the results and even if I tried and I had the data, I would probably need some training to be able to run the software. Also I did not see any description of the parameters used.
Open Source Contributions: The entire TINA toolkit is open source and it offers a nice alternative to other toolkits.
Code Quality: From what I browsed on the doxygen pages, the code is nicely written. The main point of discussion is the programing language used for the toolkit (C). Even though I am theoretically inclined to believe C++ offers much greater flexibility for software development, my experience with ITK and its highly templated code is not painless. It would be interesting to see how fast developemnt can be made by both novice and expert programmers in these two very different coding philosophies.
Tina Kapur
Monday 19 September 2005
Summary This paper introduces an open source medical image processing environment, TINA (GNU style, TINA Is No Acronym), and discusses an application that has been developed with it for measurement of cortical thickness from MR images. The goal of the project is to make both source and data available. Currently sources are available but data (for the discussed application at least) is not.
Open Source/Reproducibility
The package is open source -- sources and documentation are available on www.tina-vision.net. It is written in C (GUI in GDK/GTK, Motif) and primarily targets GNU/Linux platforms. It reportedly also compiles on MacOSX and Solaris and instructions are provided on the website for how to run it on a Windows machine using a version of linux that can run off a flash/CD drive without installing any files on the hard drive. I was able to build it under Cygwin in Windows XP (see notes below) and launched a couple of the programs mentioned in the documentation (mri_analysis and example2).
Open Science/Reproducibility I did not find either the data used by the authors in the cortical thickness measurement experiment, or documentation on which program I should invoke to compute the measures if I wanted to try it out on my own data.
Other Comments
I would recommend the README file that is distributed with the source to anyone who wants to learn more about this software. It is a very well written document both for history of the project as well as structure/compilation of the sources. Had I not been compiling in a non-certified environment, I expect the build would have been quite straightforward with the use of this file.
Based on the description of the toolkit, the user interfaces of the provided examples, as well as the documentation on the website, this package seems to be more targeted towards computer vision researchers who might be interested in duplicating/building upon the results created using TINA, rather than for clinical user who would like to, for example, meaure the cortical thickness in their own MR data.
Notes for compiling in Cygwin/Windows XP: 1)commented out AC_MINGW32 in tina-libs/configure.ac and tina-tools/configure.ac, 2) created links from cygwin libtool binary in tina-libs and tina-tools (and renamed the libtool created there),3) reinstalled x11-bin and gtk packages in cygwin. Notes for running the examples: 1)I needed to copy various cygwin dlls from their usual locations (/bin and /usr/X11R6/bin) to the directory from which I launched the examples, even though the PATH environment variable contained these locations 2)run startxwin.bat to start an x server and then ran the examples in tina-tools/toolkits/mri_analysis(example, example2)/tinaTool.exe.