IPCAS & LFCDAbout IPCASIPCAS Home

Quality Control Procedure (QCP)

A data quality control procedure (QCP) was conducted in Connectome Computation System (CCS) to ensure usable data for subsequent analyses. For structural T1 images, this QCP provides screenshots for visually inspecting the quality of image denoise, head motion, brain extraction, tissue segmentation, cortical surface reconstruction and brain registration or spatial normalization. For functional images, in addition to screenshots similar to those for the structural images, the error of nonlinear registration (errFNIRT),the minimal cost of co-registration (mcBBR) and the root mean square of frame-wise displacement (rmsFD) were computed and used to check the quality of RFMRI datasets. More details are explained below.

Anatomical preProcessing


Denoising Anatomical Image: An adaptive non-local means denoising of MR images with spatially varying noise levels (SANLM) is employed to denoise individual anatomical images. This method has been demonstrated to be able to improve the brain extraction, tissue segmentation, registration as well as subsequent resting-state functional MRI data analyses, for example, the test-retest reliability of mapping the default mode network (Zuo and Xing, 2011). SANLM
Inspecting Head Motion: A screenshot of the three images of the coronal, sagittal and axial brain is generated for users to visually inspect the obvious motion artifects during the anatomical scans. This is done based upon the SANLM denoised T1 images to avoid the potential noisy interference on the motion artifects. This step is particularly helpful for scanning kids in a MRI scanner who normally can not preserve the stable head position for a long duration. anatMotion
Extracting Brain: Several screenshots are provided for users to visually inspect the quality of brain extraction and skull stripping in CCS. Usually, CCS does good jobs on brain extraction. But, sometime, it is really bad for this purpose due to some unique brain shapes and intensity distribution (e.g., here is a bad instance). In such a situation, you can decide to exclude the subject if you have a large sample of data, or to manually intervent the pipeline in CCS and Freesurfer (finally getting a good extraction), or go with a 1-hour manual extraction of the brain if you really need this data. anatMotion
Contructing Pial/White Surfaces: A screenshot is output for users to visually inspect the quality of brain cortical surface reconstruction. There are two different colors to indicate the pial surface (yellow) and white surface (red). In most cases, the reconstruction process is always with pretty good quality if the above brain extraction is good enough of its quality. Rarely, you may need to do manual edits of these surfaces in Freesurfer software. The accuracy of constructing the pial surface is crucial to estimate the cortical thickness while that of the white surface is important for surface spatial normalization. anatMotion
Parcellating the Brain: A set of screenshots is generated by the quality assurance tools in FreeSurfer, which has been integrated into CCS. In this part, you will have a chance to visually check the quality of brain tissue segmentation, surface reconstruction as well as the final brain extraction and skull stripping. anatParcel
Spatialy Normalizing Brain: A screenshot is captured for users to assure the quality of nonlinear registration (FNIRT in FSL) of individual brains to the group template (e.g., MNI152 brain). Beyond the screenshot, the CCS pipeline also calculates the spatial correlation (r) between the normalized individual 3D image and the template image and uses the (1 -r) to measure the error of FNIRT (errFNIRT), which has been used as in (Zuo et al., 2010). You need to use this metric to exclude bad registration (e.g., errFNIRT > 0.2) if you prefer a measure in 3D image space. anatMotion

Functional preProcessing


Denoising Functional Image: An adaptive non-local means denoising of MR images with spatially varying noise levels (SANLM) is employed to denoise individual functional images. This method has been demonstrated to be able to improve the brain extraction, tissue segmentation, registration as well as subsequent resting-state functional MRI data analyses, for example, the test-retest reliability of mapping the default mode network (Xing et al., 2013). This will be included in final release of CCS. fSANLM
Inspecting Head Motion: A screenshot of mean frame-wise displacement (meanFD) and DVARS (Power et al., 2012) as well as their ovelaps in time domain was presented. In CCS, instead of meanFD, the root mean square FD (rmsFD) was used to exclude data with large head motion (rmsFD > 0.2mm) as in Patriat et al. (2013). In addition, a series of other motion metrics is produced by CCS: max displacement/rotation, time points for data scrubbing, etc. funcMotion
Extracting Functional Brain: Several screenshots are provided for users to visually inspect the quality of functional brain extraction and skull stripping in CCS. Specificlly, this procedure combines both 4D RFMRI intensity distribution and the anatomical brain extraction with BBR into a final functional brain in individual native functional space. Such a procedure is pretty important for RFMRI analyses regarding the fact that lots of processing steps (particularly the standardization, Yan et al., 2013) call global metrics as its input. fBET
Aligning Functional Images: Several screenshots are provided for users to visually inspect the quality of function-to-structure realignment. An improved GM/WM boundary based registration (BBR) algorithm is employed to achieve this purpose (Greve and Fischl, 2009). Of note, . errBBR is used to monitor the quality of function-structure co-registration, and also employed to exclude bad data (e.g., errBBR > 0.7). Of note, this threshold is dependent of the image scan quality, and Freesurfer wiki suggests 0.5 as a good indication on the quality for a 3T scanning sequence. BBR
Segmenting Functional Images: Several screenshots are provided for users to visually inspect the quality of generating white matter (WM) and CSF masks from Freesurfer. The invidual masks of WM/CSF are firstly estimated from brain tissue segmentation in Freesurfer and transformed into the individual functional space by using boundary-based registration (BBR). Users will have a chance to double check if all these masks for nuisance regression are actually in the relevant non-signal regions. MASK
Spatialy Normalizing Brain: A screenshot is captured for users to assure the quality of nonlinear registration (FNIRT in FSL) combining BBR of individual brains to the group template (e.g., MNI152 brain). Beyond the screenshot, the CCS pipeline also calculates the spatial correlation (r) between the normalized individual 3D functional image and the template image and uses the (1 -r) to measure the error of FNIRT (errFNIRT), which ahs been used as in (Zuo et al., 2010). MASK