The primary mission of the Scientific and Statistical Computing Core is to help NIH researchers with analysis of their functional MRI (brain activation mapping) data. Several levels of help are provided, ranging from short-term immediate aid to long-term development and planning.? ? Consultations:? The shortest term help consists of consultations with investigators about specific issues that arise in their research. About 180 such in-person consultations were logged in FY 2008 (including each of the NIH groups listed as NIH Collaborators), and about 3000 messages recorded at our Web-based forum. The issues that arise are quite varied, since there are many steps in carrying out an fMRI data analysis. Common problems that arise include:? - How to set up the experimental design so that the data can be analyzed effectively?? - Interpretation and correction of MRI imaging artifacts that are visible in the data (the most common of which is caused by patient head motion during scanning).? - How to set up the time series analysis to extract the brain activation effects of interest, and to suppress non-activation artifacts?? - How to analyze the data to reveal connections between brain regions during certain mental tasks, or at rest?? - How to carry out inter-patient (group) statistical analysis, especially when non-MRI data (e.g., genetic information, age, disease status) needs to be incorporated? (This is perhaps the most common class of question.)? ? Although there are familiar themes in many of these consultations, each meeting and each experiment raises unique questions and usually requires delving into the goals and details of the research project in order to ensure that nothing crucial is being missed. Complex statistical issues are often brought up, especially in the manuscript preparation phases of the users' projects.? ? Educational Efforts:? The Core has developed a 40 hour course on how to design and analyze fMRI data. This course is taught in a one week hands-on """"""""bootcamp"""""""", and was taught twice during FY 2008. All material for this continually evolving course (sample data, scripts, and PowerPoint/PDF slides) are available on our Web site http://afni.nimh.nih.gov . The course material includes several sets of sample data that are be used to illustrate the entire process, starting with images output by MRI scanners and continuing through to the collective statistical analysis of groups of subjects.? ? Algorithm and Software Development:? The longest term support consists of developing new methods and software for fMRI data analysis, both to solve immediate problems and in anticipation of new needs. Almost all of our software is incorporated into the AFNI package, which is Unix/Linux/Macintosh-based open-source and is available for download by anyone. New programs are created, and old programs modified, in response to specific user requests and in response to the Core's vision of what will be needed in the future. AFNI is """"""""pushed"""""""" to NIH computers whenever updates are made; users on non-NIH systems must manually download the software.? ? Notable developments during FY 2008 include:? - The creation and implementation of a new method for aligning functional activation data to anatomical data. We discovered that the standard and widely-used methods for such 3D image registrations would in fact fail in 5-10% of the cases. Investigating these failures in depth, we came up with a new technique that works even in very difficult cases where other software packages have trouble; our new software has a failure rate of under 1%. A paper has been submitted on this method.? - In cooperation with the NIH fMRI Facility, we implemented real-time feedback of brain activation maps to the subject in the 3 Tesla scanners at the NIH. This technique is now being used in pilot brain biofeedback studies by several groups.? - A new technique was developed and implemented to allow for non-independent noise in the brain mapping time series data. Other methods widely used to allow for this statistical problem assume that the time series noise has the same mathematical structure at every point in the brain. This assumption is known to be wrong, but is made for the sake of computational efficiency. Our new implementation overcomes this problem, and corrects the statistical estimates of the brain activation maps separately for each part of the brain. Our method is actually faster than most software that uses the cruder global correction technique.? - A new statistical technique was implemented for combining individual subject brain activity maps into group maps. This software allows for intra-subject correlations in the brain maps, and can deal with several other problems that also commonly arise: unbalanced numbers of subjects in different groups, allowance for non-brain subject-specific data (e.g., age, IQ), and missing data points.? - Two new techniques for brain connectivity analysis were implemented in software this year. They are currently being tested by users. The initial reports on our """"""""vector autogressive causality"""""""" software are very promising.? ? In addition, many small-to-medium changes were made to the software in response to specific NIH researcher requests and needs. And (of course) many small bug fixes were made -- we pride ourselves on fixing bugs in AFNI within a few days of their report.? ? Our future development plans include the creation and implementation of a semi-linear global deconvolution method for brain time series analysis, which will provide a more realistic model for the brain data than is currently available in any widely used software package. We also plan a major effort to improve our software for resting state brain connectivity analysis, in cooperation with Dr Alex Martin (LBC/NIMH). This type of data is hoped to be important for understanding the brain activity of patients who cannot easily carry out complex tasks in an MRI scanner, including autism and Alzheimer's patients.? ? Extramural Collaborations, etc.:? - In FY 2008, we incorporated Dr LR Frank's (UCSD) complex high angular resolution diffusion imaging (HARDI) software into AFNI;? - We incorporated into AFNI more brain atlas databases developed by Dr K Zilles (Julich);? - We fully incorporated Dr SM Laconte's (Baylor) brain-state classifier software into AFNI;? - We took the lead in the GIFTI initiative to develop and implement a standard data format for exchange of brain cortical surface model--Dr Saad was the leader of the GIFTI consortium, and Mr Reynolds developed the open-source software for GIFTI;? - Approximately half of all AFNI message board traffic is from extramural AFNI users. On a space-available basis, the SSCC also allows a few non-NIH users to sit in on the bi-annual AFNI bootcamps at the NIH.? ? As a measure of our impact, thus far in CY 2008 (Jan-Sep), the principal AFNI publication has been cited in 159 papers: 22 from the NIH, and the rest from extramural institutions. Over 30 such institutions had 3 or more such papers citing AFNI, indicating the breadth of our contribution to brain imaging research. (Figures from Science Citation Index online.)
Showing the most recent 10 out of 13 publications
