neurobureau:NIAKPipeline

From NITRC Wiki

Contents 1 Overview of the NIAK preprocessing release of ADHD200 2 ADHD200 Phenotypic data 3 Quality control of the preprocessing - Training dataset 3.1 KKI 3.1.1 T1-fMRI coregistration 3.1.2 T1-stereotaxic space coregistration 3.1.3 Motion 3.2 OHSU 3.2.1 T1-fMRI coregistration 3.2.2 T1-stereotaxic space coregistration 3.2.3 Motion 3.3 Beijing 3.3.1 T1-fMRI coregistration 3.3.2 T1-stereotaxic space coregistration 3.3.3 Motion 3.4 Pittsburgh 3.4.1 T1-fMRI coregistration 3.4.2 T1-stereotaxic space coregistration 3.4.3 Motion 3.5 NYU 3.5.1 T1-fMRI coregistration 3.5.2 T1-stereotaxic space coregistration 3.5.3 Motion 3.6 Neuroimage 3.6.1 T1-fMRI coregistration 3.6.2 T1-stereotaxic space coregistration 3.6.3 Motion 3.7 Washington University 3.7.1 T1-fMRI coregistration 3.7.2 T1-stereotaxic space coregistration 3.7.3 Motion 4 Quality control of the preprocessing - Test dataset 4.1 KKI 4.2 Pittsburgh 4.3 OHSU 4.4 Peking 4.5 Brown 4.6 NYU 4.7 Neuroimage 5 Functional parcelation 5.1 Method 5.2 Results 5.3 References

Overview of the NIAK preprocessing release of ADHD200

The ADHD200 has been preprocessed with the pipeline available as part of the Neuroimaging Analysis Kit (NIAK) for Octave and Matlab. Links for download are available on the NITRC Neurobureau website. A full description of this initiative can be found in the following paper, please cite this work if you use the NIAK preprocessed ADHD-200 sample in a publication:

Lavoie-Courchesne, S., Rioux, P., Chouinard-Decorte, F., Sherif, T., Rousseau, M. E., Das, S., Adalat, R., Doyon, J., Craddock, C., Margulies, D., Chu, C., Lyttelton, O., Evans, A. C., Bellec, P., Feb. 2012. Integration of a neuroimaging processing pipeline into a pan-canadian computing grid. Journal of Physics: Conference Series 341 (1), 012032+. URL http://dx.doi.org/10.1088/1742-6596/341/1/012032

The tools used to preprocess the datasets are based on the MINC tools and the processing was done through the CBRAIN platform for grid computing. The main qualitative differences between the NIAK release and the Athena release is that (1) NIAK uses an automated detection of physiological noise components in an ICA (CORSICA) rather than a regression approach, (2) no low pass filtering, and, (3) the resolution of functional volumes in stereotaxic space is 3 mm isotropic rather than 4 mm. The flowchart of the pipeline is as follows :

1. Slice timing correction (piecewise cubic spline temporal interpolation) - NIAK_BRICK_SLICE_TIMING
2. Motion correction (rigid-body, to the median volume of the resting-state of each run, and then between-runs/sessions) - NIAK_PIPELINE_MOTION_CORRECTION
3. Quality control for motion correction. 
4. Linear and non-linear spatial normalization of the anatomical image (and many more anatomical stuff such as brain masking and CSF/GM/WM classification, see below) - NIAK_BRICK_T1_PREPROCESS
5. Coregistration of the anatomical volume with the mean functional volume. - NIAK_BRICK_ANAT2FUNC
6. Concatenation of the T2-to-T1 and T1-to-stereotaxic-nl transformations.
7. Extraction of mean/std/mask for functional images, in various spaces (Linear and non-linear stereotaxic spaces).
8. Quality control for 4 and 5.
9. Correction of slow time drifts (high-pass filtering with discrete cosines, cut-off frequency 0.01Hz). - NIAK_BRICK_TIME_FILTER
10. Correction of physiological noise (20 components, selection threshold 0.15) - NIAK_PIPELINE_CORSICA
11. Resampling of the functional data in the stereotaxic space (tricubic spatial interpolation, 3mm isotropic resolution). - NIAK_BRICK_RESAMPLE_VOL
12. Spatial smoothing (Gaussian kernel, 6mm isotropic). - NIAK_BRICK_SMOOTH_VOL

The flowchart of the processing of the T1 image is more specifically :

1. Non-uniformity correction in native space (without mask) - NIAK_BRICK_NU_CORRECT
2. Brain extraction in native space - NIAK_BRICK_MASK_BRAIN_T1
3. Linear coregistration in stereotaxic space (with mask from 2) - NIAK_BRICK_ANAT2STEREOLIN
4. Non-uniformity correction based on the template mask - NIAK_BRICK_NU_CORRECT
5. Brain extraction, combined with the template mask - NIAK_BRICK_MASK_BRAIN_T1
6. Intensity normalization - NIAK_BRICK_INORMALIZE
7. Non-linear coregistration in template space (with mask from 5) - NIAK_BRICK_ANAT2STEREONL
8. Generation of the brain mask in the non-linear stereotaxic space by intersection of the template mask with a head mask. - NIAK_BRICK_MASK_HEAD_T1, NIAK_BRICK_MATH_VOL
9. Generation of the mask in the stereotaxic linear space by application of the inverse non-linear transform from 7 and the brain mask from 8. - NIAK_BRICK_RESAMPLE_VOL
10. Tissue classification - NIAK_BRICK_CLASSIFY

Note that some parameters (slice timing scheme, scanner type, and parameter for non-uniformity correction) were selected on a site-by-site basis. The preprocessed fMRI resting-state datasets will be made available along with various results derived from the T1 image (including the non-linear deformation grid in stereotaxic space), as well as measures for quality control purposes. Summaries of quality control for each site will be posted here as well. For more details regarding the NIAK preprocessing pipeline, please refer to the pdf user guide and the NIAK fMRI preprocessing tutorial.

Each site will have four tar.gz archives released:

• The preprocessed fMRI resting-state datasets (fmri_<site>.tar).
• The preprocessed anatomical (T1) volumes (anat_<site>.tar).
• The intermediate results (intermediate_<site>.tar). This will notably include the motion parameters in .mat files.
• The quality control results (quality_control_<site>.tar). A brief summary for each site can be found below.

The NIAK wiki includes a short description of the pipeline for publication purposes, along with a list of suggested references.

ADHD200 Phenotypic data

Phenotypic data for each site can be downloaded from the download section: download section or from the following links: training set phenotypic datatesting set phenotypic data. A key describing the different data measures can be found here.

Quality control of the preprocessing - Training dataset

KKI

T1-fMRI coregistration

Initially, the T1-fMRI failed for four subjects (X_1577042, X_2344857, X_8628223, X_3699991). The pipeline was restarted on these subjects only, after selecting an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). This solved the problem. The results of these four subjects have been updated for the release, except for the QC results which correspond to the first pass.

T1-stereotaxic space coregistration

QC only one subject with a problem in the non-linear coregistration in stereotaxic space: X_2740232. This subject has a very asymmetric head, and the non-linear coregistration algorithm has pushed part of the skull into the occipital cortex. Because the coregistration on the rest of the brain is satisfactory, this subject may still be included in an analysis.

Motion

Only one subject has signs of severe motion (X_3103809) with over 3 degrees in translation of rotation parameters. Otherwise the following subjects had a slightly high amount of motion (over 1 mm transtition in translation, or 1 degree in transition in rotation) : X_2703289, X_1541812, X_2299519, X_1846346, X_2138826, X_1577042, X_1686265, X_1962503, X_2360428, X_8432725, X_3310328 Those can still be included in an analysis. The remaining subject had negligible motion (less than 1 mm transtition in translation, or 1 degree in transition in rotation).

OHSU

T1-fMRI coregistration

The T1-fMRI coregistration failed for four subjects (X_3358877, X_3812101, X_4072305, X_4529116). The pipeline was restarted on these subjects only, after selecting an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). The new registration was satisfactory and the results of the four subjects have been updated for the release.

T1-stereotaxic space coregistration

Six subjects showed sub-standard quality of the T1 scans that resulted in poor normalization to stereotactic space (X_2292940, X2535204, X_2920716, X_1743472. X_1536593, X_3560456). This is likely attributable to excessive movement and we advise excluding these subjects from the analyses.

Motion

Of the 79 subjects, 44 showed minimal movement. 19 subjects had moderate movement of 1 to 3mm in translation and 1-3 degrees in rotation and can still be considered for inclusion in the analyses (X_4529116, X_3812101, X_3652932, X_2426523, X_2054310, X_2124248, X_8064456, X_3466651, X_4072305, X_2790141, X_4219416, X_3869075, X_2929195, X_3470141, X_4103874, X_2561174, X_7333005, X_2409220, X_3206978). 16 subjects showed movement exceeding 3mm  translation or 3 degrees rotation and should be excluded from analysis (X_3560456, X_2455205, X_1386056, X_2288903, X_3286474, X_1696588, X_2620872, X_3684229, X_8720244, X_2920716, X_2559559, X_2054998, X_1536593, X_2845989, X_2292940, X_2535204).

Beijing

T1-fMRI coregistration

Three subjects (X_2919220, X_2268253, X_2950754) were missing dorsal slices and should be excluded from analysis. This was found manually on a subset of subjects and more might have the same problem. The T1-fMRI coregistration failed for two subjects (X_1805037, X_4053836). The pipeline was restarted on these two subjects only, after selecting an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). The new registration was satisfactory and the results of the two subjects have been updated for the release.

T1-stereotaxic space coregistration

One subject (X_3291029) showed an acquisition artifact and should be excluded from analysis.

Motion

Of the 194 subjects, 175 showed minimal movement. 14 subjects had moderate movement of 1 to 3mm in translation and 1-3 degrees in rotation and can still be considered for inclusion in the analyses (X_2529026, X_3205761, X_4334113, X_2107404, X_2174595, X_2228148, X_1805037, X_1494102, X_9002207, X_3593327, X_3561920, X_4073815, X_1947991, X_3993793, ). 5 subjects showed movement exceeding 3mm  translation or 3 degrees rotation and should be excluded from analysis (X_1791543, X_4225073, X_2296326, X_1860323, X_2367157).

Pittsburgh

T1-fMRI coregistration

The T1-fMRI registration worked well for all subjects. Subject X_0016029 was found to be missing some dorsal slices in the fMRI data and other subjects in the dataset might also share this problem.

T1-stereotaxic space coregistration

The non-linear coregistration in stereotaxic space of the T1 images worked well for all subjects.

Motion

Of the 89 subjects 65 showed minimal movement. 14 had movement ranging from 1 to 3 mm (X_0016069, X_0016009, X_0016028, X_0016005, X_0016083, X_0016034, X_0016048, X_0016044, X_0016013, X_0016054, X_0016041, X_0016047, X_0016053, X_0016003). 10 subjects had movement exceeding 3mm and should excluded from analysis (X_0016072, X_0016040, X_0016024, X_0016023, X_0016025, X_0016015, X_0016026, X_0016017, X_0016007, X_0016079).

NYU

T1-fMRI coregistration

The T1-fMRI coregistration initially failed for thirty-six subjects. The pipeline was restarted on these subjects only, after selecting an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). This worked for all subjects and the results have been updated for the release (QC results correspond to the first pass).

T1-stereotaxic space coregistration

Twelve subjects showed poor normalization to stereotactic space. The pipeline was restarted with 'Nu correct distance = 100' for these subjects. This corrected the problems for five subjects and the remaining seven should be excluded from analyses (X_0010013, X_2297413, X_3679455, X_0010032, X_1435954, X_6206397, X_8415034).

Motion

Of the 216 subjects 136 showed minimal movement. 55 had movement ranging from 1 to 3 mm (X_0010032, X_4562206, X_0010074, X_0010086, X_4079254, X_3542588, X_0010012, X_1992284, X_9578663, X_1700637, X_2741068, X_2030383, X_3433846, X_2107638, X_4084645, X_0010018, X_1780174, X_0010031, X_0010011, X_0010102, X_3619797, X_0010060, X_0010050, X_1497055, X_3235580, X_3243657, X_0010041, X_1000804, X_1854959, X_2230510, X_1918630, X_0010090, X_0010044, X_0010028, X_2054438, X_0010049, X_3845761, X_0010096, X_0010047, X_0010019, X_2136051, X_2821683, X_0010022, X_0010033, X_3163200, X_0010062, X_2735617, X_0010092, X_2907383, X_3349205, X_2950672, X_5971050, X_0010106, X_3653737, X_0010085).25 subjects had movement exceeding 3mm and should excluded from analysis (X_8915162, X_0010061, X_0010025, X_0010117, X_6568351, X_3518345, X_3601861, X_4187857, X_0010015, X_0010014, X_0010078, X_0010030, X_1187766, X_0010046, X_0010095, X_0010077, X_0010005, X_3662296, X_0010119, X_0010103, X_0010003, X_0010066, X_0010111, X_0010067, X_1023964).

Neuroimage

T1-fMRI coregistration

The T1-fMRI coregistration initially failed for a majority of subjects and the pipeline was restarted with an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). This worked for all subjects and the results and QC have been updated for the release.

T1-stereotaxic space coregistration

The normalization to stereotactic space initially failed for a number of subjects and was restarted with the option 'Nu correct distance = 100'. This corrected most problems and the coregistration failed for only one subjects (3304956) which should be excluded from the analyses.

Motion

Of the 48 subjects 37 showed minimal movement. 6 had movement ranging from 1 to 3 mm (1580708, 1312097, 3108222, 4239636, 1438162, 5045355). Five subjects had movement exceeding 3mm and should excluded from analysis (2029723, 3048588, 3082137, 4919979, 3808273).

Washington University

T1-fMRI coregistration

The T1-fMRI coregistration initially failed for three subjects and the pipeline was restarted with an initial realignement of the center of mass of the brain masks in T1 and fMRI (opt.anat2func.init = 'center'). This worked for all subjects and the results and QC have been updated for the release.

T1-stereotaxic space coregistration

The non-linear coregistration in stereotaxic space of the T1 images worked well for all subjects.

Motion

Of the 49 subjects 23 showed minimal movement. 17 had movement ranging from 1 to 3 mm (X_0015012, X_0015021, X_0015025, X_0015029, X_0015035, X_0015055, X_0015050, X_0015022, X_0015002, X_0015044, X_0015049, X_0015037, X_0015047, X_0015008, X_0015003, X_0015051, X_0015056). Nine subjects had movement exceeding 3mm and should excluded from analysis (X_0015034, X_0015015, X_0015020, X_0015010, X_0015060, X_0015059, X_0015038, X_0015046, X_0015023).

Quality control of the preprocessing - Test dataset

KKI

All subjects passed quality control (T1-fMRI as well as T1-stereotaxic space coregistration, and motion correction).

Pittsburgh

All subjects passed quality control (T1-fMRI as well as T1-stereotaxic space coregistration, and motion correction).

OHSU

All subjects passed the registration quality control (T1-fMRI as well as T1-stereotaxic space coregistration).Of the 34 subjects 28 showed minimal movement. Six subjects had movement between 1 and 3mm : 0023011, 0023027, 0023030, 0023033, 0023007, 0023024.

Peking

All subjects passed the registration quality control (T1-fMRI as well as T1-stereotaxic space coregistration). Of the 50 subjects 46 showed minimal movement. Four subjects had movement between 1 and 3mm : 1245758, 3169448, 2411995, 3378296.

Brown

All subjects passed the T1-fMRI registration quality control and one subject showed problems in the T1-stereotaxic space coregistration (X_0026016) we advise excluding this subject from the analyses. Of the 26 subjects 24 showed minimal movement. Four subjects had movement between 1 and 3mm : 0026030, 0026054.

NYU

Four subjects failed the T1-fMRI registration quality control and three subject failed the T1-stereotaxic space coregistration. The T1-fMRI registration pipeline was successfully restarted on all subjects. The T1-stereotaxic space coregistration pipeline was restarted unsuccessfully on the three subjects and we advise excluding them from the analyses (0021046, 0021014, 0021037). All subjects had movement under 1mm.

Neuroimage

All subjects passed the registration quality control (T1-fMRI as well as T1-stereotaxic space coregistration). Of the 25 subjects 23 showed minimal movement and only two subjects had movement between 1 and 3mm : 0027007, 0027042.

Functional parcelation

Method

In order to limit the computational burden of subsequent analysis, a region-growing algorithm (Bellec et al. 2006) was applied to the concatenated time series of every subjects in order to derive a common segmentation of the brain into small functionally homogeneous regions. This procedure was based on a single site (KKI), excluding the subject with large motion (X_3103809) yet including the subject with deformation in the occipital cortex (X_2740232). The remaining sites can be used to train/test a classifier without introducing any circularity in the analysis (the functional parcels are defined independently of the test set). Moreover, to limit the memory demand, the region-growing was applied independently in each of the 116 areas of the AAL template (Tzourio-Mazoyer et al., 2002). The resulting regions were spatially connected with roughly equal size, which was set to 1000 mm3 and 330 mm3, translating into roughly 1000 and 3000 regions respectively covering the grey matter. The average time series within each region was derived after correction to a zero temporal mean and unit variance, and then concatenated across all the fMRI datasets for each subject.

Results

The following table summarizes the size profile of the generated functional parcelations (in mm3) :

summary of size parameters for the functional parcelations

	size threshold	# of ROIs	mean size	std size	min size	max size
rois_1000	1000	954	1404	365.72	27	2781
rois_3000	330	2843	471	108.88	27	1026

The following figures represent the distribution of the size of the parcels (in mm3):

The whole-brain connectome averaged across the subjects from KKI using rois_1000 is represented below. This graph was generated using gephi, a threshold of 0.3 on the correlation matrix, and an Yfan Hu's multi-level layout. Color and size code for the degree of each node (bluer = larger). Thanks to Cameron Craddock for pointing out gephi.

References

P. Bellec, et al. (2006). `Identification of large-scale networks in the brain using fMRI'. NeuroImage 29(4):1231-1243.

N. Tzourio-Mazoyer (2002). `Automated Anatomical Labeling of Activations in SPM Using a Macroscopic Anatomical Parcellation of the MNI MRI Single-Subject Brain'. NeuroImage 15(1):273-289.