Skip to content

Latest commit

 

History

History
223 lines (172 loc) · 13.8 KB

README_pandoc.md

File metadata and controls

223 lines (172 loc) · 13.8 KB

Awesome MRI QC

A curated list of tools, measures, and references for MRI quality control (QC).

Quality Control (QC) refers to “a real-time prospective process to ensure imaging quality is maintained by comparing it regularly to a defined set of criteria or industry standards” [@Sreedher2021] via [@Das2022].

Inspired by awesome-python and other awesome lists.

Tools

QC metric and report generation

  • MRIQC: extracts no-reference image quality metrics from structural and functional MRI data.
  • QAP: The QAP package allows you to obtain spatial and anatomical data quality measures for your own data. (Precursor to MRIQC.)
  • fBIRN QA tools: These tools form the basis of the fBIRN QA procedures [@Glover2012].
  • MRQy: A quality assurance and checking tool for quantitative assessment of MRI data

MRI analysis packages with built-in QC

  • AFNI: A suite of programs for looking at and analyzing MRI brain images at all stages of analysis.
    • 3dToutcount: Calculates number of "outliers" a 3D+time dataset, at each time point.
    • 3dTqual: Computes a "quality index" for each sub-brick in a 3D+time dataset.
  • FSL: A comprehensive library of analysis tools for FMRI, MRI and DTI brain imaging data.
    • EDDY QC tools : Generates single-subject and group QC reports for diffusion MRI.
    • FEAT report: HTML report for single-subject and group fMRI analysis.
    • fsl_motion_outliers: command-line tool for analyzing single-subject motion.
  • NiPype: Provides a uniform interface to existing neuroimaging software and facilitates interaction between these packages within a single workflow
  • C-PAC: A configurable, open-source, Nipype-based, automated processing pipeline for resting state functional MRI data
  • XCP: A free, open-source software package for processing of multimodal neuroimages with extensive QC metrics for T1w and BOLD MRI images.
  • DPABI: A toolbox for Data Processing & Analysis for Brain Imaging including GUI-based QC reports.
  • DSI Studio: A tractography software tool that maps brain connections and correlates findings with neuropsychological disorders. Includes automated QC metrics.
  • QSIprep: Configures pipelines for processing diffusion-weighted MRI (dMRI) data. Includes QC metrics from DSI Studio.

QC at scale

  • NiRV: A modern neuroimaging report viewer that aggregates participant level HTML reports for datasets, small and large.
  • NiReports: The NiPreps' Reporting and Visualization system - report templates and "reportlets"
  • SQAN: Scalable Quality Assurance for Neuroimaging. A full-stack system for extracting, translating, logging, and visualizing DICOM-formatted medical imaging data.

QC web platforms

  • MIQA: Efficient and accurate QC processing by leveraging modern UI/UX and deep learning techniques
  • MindControl: An app for quality control of neuroimaging pipeline outputs, especially anatomical segmentations
  • Braindr: a firebase app for braindr: Tinder for brains
  • Fibr: An app for quality control of diffusion MRI images from the Healthy Brain Network
  • dmriprep-viewer: Web app to visualize local QSIprep and dMRIprep outputs

Automated QC with machine learning

  • Qoala-T: Qoala-T is a supervised-learning tool for quality control of FreeSurfer segmented MRI data
  • mriqc-learn: Learning on MRIQC-generated image quality metrics (IQMs)

General-purpose image quality estimation

Measures

Structural T1

Image quality

Measure Summary Interpretation References
Coefficient of joint variation (CJV) Larger values indicate head motion and INU artifacts lower better MRIQC, [@Ganzetti2016]
Contrast-to-noise ratio (CNR) Larger values indicate more GM to WM contrast higher better MRIQC, [@Magnotta2006]
Signal-to-noise ratio (SNR) SNR within brain mask higher better MRIQC
Dietrich SNR (SNRd) SNR relative to air background higher better MRIQC, [@Dietrich2007]
Mortamet’s quality index 1 (QI1) Proportion of "corrupted" voxels vs number of background voxels lower better MRIQC, [@Mortamet2009]
Mortamet’s quality index 2 (QI2) Comparison of background noise with $\Chi^2$ distribution after correcting for QI1 lower better MRIQC, [@Mortamet2009]
EFC Shannon entropy as indicator of ghosting due to head motion lower better MRIQC, [@Atkinson1997]
FBER Ratio of "mean energy" within head vs air higher better MRIQC, [@Shehzad2015]
INU Summary stats for INU bias field. Values close to 1 mean less bias. higher better MRIQC, [@Tustison2010]
White matter to maximum ratio (WM2max) Median WM intensity divided by 95 percentile intensity values in [0.6, 0.8] are good. MRIQC
FWHM Estimation of image smoothness. Higher values mean blurry. lower better MRIQC

Tissue segmentation

Measure Summary Interpretation References
ICV Intracranial volume fraction for each tissue type (WM, GM, CSF) MRIQC
rPVe Residual partial voluming error for each tissue type lower better MRIQC
Tissue summary stats Summary stats for signal within tissue masks (mean, stdev, p05, p95) MRIQC
Tissue prior overlap Overlap of estimated tissue probability maps with template priors higher better MRIQC
Tissue skewness/kurtosis Skewness and kurtosis of intensity distribution for WM, GM, and background MRIQC, [@Rosen2018]
WM hypointensities Voxel count of white matter hypointensities lower better Freesurfer, [@Klapwijk2019]

Brain extraction

Measure Summary Interpretation References
Boundary tissue count Volume of each tissue type lying on brain mask boundary. Large WM values suggest brain extraction failure. lower WM values better [@Alfaro-Almagro2018]

Spatial normalization

Measure Summary Interpretation References
Normalization cost Cost function between T1 and template under linear and nonlinear alignment lower better
Normalization magnitude Amount of nonlinear warping lower better [@Alfaro-Almagro2018]
Normalized overlap Dice or Jaccard overlap coefficient between resampled T1 and template for brain and tissue masks higher better XCP, [@Alfaro-Almagro2018]

Surface reconstruction

Measure Summary Interpretation References
Euler number 2 - 2g where g is the number of topological holes in the surface. Computed by the Euler–Lhulier formula (V - E + F) higher better [@Rosen2018], [@Dale1999], Freesurfer
Local gyrification index (LGI) Measures degree of cortical folding in neighborhood of each vertex (spatial map). Freesurfer
BBR criterion Measures the magnitude of WM/GM contrast across the WM surface boundary higher better [@Greve2009], Freesurfer

Cortical/subcortical segmentation

Measure Summary Interpretation References
Subcortical volume Volume of each subcortical region in a segmentation (vector). FIRST, Freesurfer
Cortical volume Volume of each cortical region in a parcellation (vector). [@Alfaro-Almagro2018], ANTs Freesurfer, Mindboggle
Cortical thickness Mean thickness of each region in a parcellation (vector). [@Rosen2018], ANTs, Freesurfer

Functional BOLD

Image quality

Measure Summary Interpretation References
EFC Shannon entropy as indicator of ghosting due to head motion lower better MRIQC, [@Atkinson1997]
FBER Ratio of "mean energy" within head vs air higher better MRIQC, [@Shehzad2015]
FWHM Estimation of image smoothness. Higher values mean blurry. MRIQC
SNR SNR within brain mask. higher better MRIQC
BOLD summary stats BOLD intensity summary stats (mean, stdev, p95, p05) MRIQC
Global correlation (GCor) Average correlation between every voxel and every other voxel AFNI, MRIQC, [@Saad2013]
Temporal standard deviation (tSD) Map of temporal standard deviation lower better MRIQC, [@Marcus2013]
Temporal SNR (tSNR) Map of temporal mean divided by standard deviation higher better MRIQC
Ghost to signal ratio (GSR) Measures amount of signal in regions prone to ghosting lower better MRIQC
AFNI outlier ratio (AOR) Mean fraction of "outliers" per fMRI volume using AFNI 3dToutcount lower better AFNI, MRIQC
AFNI quality index (AQI) Mean "quality index", which for each volume is 1 - correlation to median volume lower better AFNI, MRIQC
Number of dummy scans Number of non-steady state dummy scans MRIQC
Carpet plot BOLD time series for a set of ROIs arranged in a matrix MRIQC, [@Power2017]
Air signal mean BOLD time series for a set of background/air slices MRIQC

Motion correction

Measure Summary Interpretation References
DVARS Measures amount of signal change between consecutive time points (time series) spikes indicate significant motion MRIQC, NiPype, fsl_motion_outliers, [@Power2012]
Framewise displacement (FD) Sum of absolute translation and rotation displacements in mm at each time point (time series) spikes indicate significant motion MRIQC, NiPype, [@Jenkinson2002], [@Power2012]

Co-registration

Measure Summary Interpretation References
Co-registration cost Cost function for rigid registration between BOLD and T1 lower better XCP
Brain mask overlap Dice or Jaccard brain mask overlap coefficient between resampled BOLD and T1 higher better XCP

Functional connectivity

TODO

Diffusion weighted imaging (DWI)

Image quality

Measure Summary Interpretation References
Mean neighbor correlation Average Pearson correlation between each diffusion image and its q-space nearest neighbor expected range [0.6, 0.8] QSIprep, DSI Studio, [@Yeh2019]
Dropout slice count Count of slices with significant signal dropout expected less than 0.1% QSIprep, DSI Studio, [@Yeh2019]
Fiber coherence index Measures how well fibers are connected to each other low values indicate flipped b-vectors QSIprep, DSI Studio, [@Schilling2019]

Fiber orientation modeling

TODO

Tractography

TODO

Structural connectivity

TODO

References