AFNI Robert W Cox, PhD National Institute of Mental Health Bethesda, MD afni.nimh.nih.gov/afni http:// afni.nimh.nih.gov/afni.

AFNIAFNIRobert W Cox, PhDRobert W Cox, PhD

National Institute of Mental Health

Bethesda, MD

http://afni.nimh.nih.gov/afniafni.nimh.nih.gov/afni

An FMRI Analysis An FMRI Analysis EnvironmentEnvironment Philosophy:

Encompass all needed classes of data and computations

Extensibility + Openness + Scalability:Anticipating what will be needed to solve problems that have not yet been posed

Interactive vs. Batch operations: Stay close to data or view from a distance

Components: Data Objects: Arrays of 3D arrays + auxiliary data Data Viewers: Numbers, Graphs, Slices, Volumes Data Processors: Plugins, Plugouts, Batch Programs

Steps in Processing with Steps in Processing with AFNIAFNI Image assembly into datasets [to3d ]

Can be done at scanner with realtime plugin

Image registration [3dvolreg] Functional activation [AFNI, 3dfim]

Linear and nonlinear time series regression [3dDeconvolve, 3dNLfim]

Transformation to Talairach [AFNI ] Or: selection of anatomical ROIs [AFNI ]

Integration of results across subjects [many] Visualization & thinking [AFNI & you]

AFNIAFNI Controller Window Controller Window

Interactive Analysis with Interactive Analysis with AFNIAFNI

Graphing voxeltime series data

Displaying EP imagesfrom time series

ControlPanel

FIM overlaid on SPGR, in Talairach coords

Multislice layouts

Looking at the Results

Volume Rendering ControlsVolume Rendering Controls

SampleRendering:

Coronal sliceviewed from side;

function not cut out

Rendering is easy tosetup and carry outfrom control panel

“ShowThru” Rendering of Function

Integration of ResultsIntegration of Results Done with batch programs (usually in scripts) 3dmerge: edit and combine 3D datasets 3dttest: voxel-by-voxel t-tests 3dANOVA:

Voxel-by-voxel: 1-, 2-, and 3-way layouts Fixed and random effects Other voxel-by-voxel statistics are available

3dpc: principal components (space time) ROI analyses are labor-intensive alternative

Extending Extending AFNIAFNI Package Package Batch programs

Output new 3D datasets for viewing with AFNI

Plugins — searched for & loaded at startup Add interactive capabilities to AFNI program “Fill in the blanks” menu for input from users 40 page manual and some samples included

Plugouts — attach themselves during run External programs that communicate with AFNI

with shared memory or TCP/IP sockets

Whole Brain Whole Brain Realtime FMRIRealtime FMRI

Assembly of images into AFNI datasets during acquisition Can use AFNI tools to visualize during scanning

Realtime 3D registration Graph of estimated motion parameters

Recursive signal processing to update activation map with each new data volume Color overlay changes with each TR

Realtime Realtime AFNIAFNI AFNI software package has a realtime

plugin, distributed with every copy Price: USD$0 [except for time & effort] Runs on Unix/Linux Requires input of reconstructed images and

geometrical information about them

For more information see Web site

The Goal: Interactive The Goal: Interactive Functional Brain MappingFunctional Brain Mapping

See functional map as scanning proceeds

1 minute 2 minutes 3 minutes

Registration Goals for Registration Goals for Online FMRIOnline FMRI

Estimate 3D (6 DOF) movement parameters as fast as volume acquisition happens

Realign each volume to a “target” volume during scanning

Display updating graphs of estimated motions to investigator

Feedback movements to slice selection?

EstimatedEstimatedsubjectsubjectmovementmovementparametersparameters

Multiple ReferencesMultiple References v(t) 1 h1(t ) + 2 h2(t ) +

1 , 2 are amplitudes (unknown)

h1 , h2 are known reference responses

Used for experiments with more than 1 stimulus condition:

rest task A rest task B rest task A

h1 0 h2 0 h1 0

Widely used for event-related FMRI

Linear DeconvolutionLinear Deconvolution v(t) a + bt + j h(tj) + noise

j = stimulation times [known]

h(t) = k kuk(t) = response function

uk(t) = basis functions [known]

k = amplitudes [unknown]

Goal is to find shape and amplitude of response function in each voxel Unlike previous analyses, form of response is not

completely bound to hemodynamic assumptions

Recursive Linear RegressionRecursive Linear Regression All methods above can be cast into form of

linear regression: Solution of linear equations to get estimated fit

parameters Estimation of significance from noise model

[i.e., using what’s left after regression fit] Recursive regression:

With each new time point, add one equation Given previous solution, can re-compute new fit

with relatively little work (much less than starting over)

Method used in AFNI for realtime analysis

v(t) a + bt + j h(tj) + noise

h(t) = h(t ; ) = nonlinearly dependent on

= vector of unknown parameters Example: h(t) = A (t)r exp((t)/c)

Nonlinear Regression orNonlinear Regression orDeconvolutionDeconvolution

0 1 2 3 4 5 6 7 8 9 10 11

r = 8.6 c = 0.54 = 0

Single Event FMRI:Single Event FMRI:Use Nonlinear Regression?Use Nonlinear Regression?

In this type of experiment, the stimulus and its consequences last a long time

Only have one stimulus/response event per imaging run Administration of a drug Presentation of an affect altering video

Know when stimulus started, but don’t know exactly what response should be Nonlinear curve fitting seems appropriate