Visual QC Protocol for FreeSurfer Cortical Parcellations from Anatomical MRI

Abstract

Quality control of morphometric neuroimaging data is essential to improve reproducibility. Owing to the complexity of neuroimaging data and, subsequently, the interpretation of their results, visual inspection by trained raters is the most reliable way to perform quality control. Here, we present a protocol for visual quality control of the anatomical accuracy of FreeSurfer parcellations, based on an easy to use open source tool called VisualQC. We comprehensively evaluate its utility in terms of error detection rate and inter-rater reliability on two large multi-site datasets and discuss and demonstrate site differences in error patterns. This evaluation shows that VisualQC is a practically viable protocol for community adoption.

Automatic quality control of brain T1-weighted magnetic resonance images for a clinical data warehouse

“…IQMs are then compared to a normative distribution obtained from three research datasets, ABIDE (Di Martino et al, 2014), CoRR 2 and NFB 3 . In the same spirit, we find (Esteban et al, 2017;Alfaro-Almagro et al, 2018;Raamana et al, 2020). These approaches propose to use the IQMs as input of a classifier for automatic QC.…”

“…MRIQC (Esteban et al, 2017) and VisualQC (Raamana et al, 2020) are two tools developed for the QC of T1w brain MRI data: they propose the extraction of image quality metrics for the detection of outliers, and a graphical interface to check the images. (Alfaro-Almagro et al, 2018) proposed a pipeline for the UK Biobank dataset.…”

“…(Esteban et al, 2017) and (Alfaro-Almagro et al, 2018) developed a pipeline for the automatic QC of 3D brain T1w MRI, the first has the advantage to be an open source software (called MRIQC). (Raamana et al, 2020) developed another open source software called VisualQC whose aim is the visualisation and the rating of the Freesurfer cortical segmentation output. The pipelines proposed by these works are very extensive as they require registration and segmentation steps to extract features.…”

The Open-Access European Prevention of Alzheimer’s Dementia (EPAD) MRI dataset and processing workflow

“…The application of machine learning classifiers has proved its efficiency in recent QC efforts on classifying 3D T1w image-quality from QC features distribution, both the UKBiobank (F. Alfaro-Almagro and Jenkinson 2016 ) and atasets. Similarly, random forest classifiers trained on FreeSurfer QC output showed good accuracy in scanning site identification, supporting the use of multivariate approaches for QC metrics’ importance evaluation ( Raamana et al, 2021 ). However, while these works aimed at the fully automatic prediction of image quality from unseen scans/sites, we focused on identifying a set of informative QC features as a pre-selection and guide for visual inspection.…”

The European Prevention of Alzheimer’s Dementia (EPAD) MRI Dataset and Processing Workflow

“…The application of machine learning classifiers has proved its efficiency in recent QC efforts on classifying 3D T1w image-quality from QC features distribution, both the UKBiobank (F. Alfaro-Almagro and Jenkinson 2016) and ABIDE Di Martino et al 2014) datasets. Similarly, random forest classifiers trained on FreeSurfer QC output showed good accuracy in scanning site identification, supporting the use of multivariate approaches for QC metrics' importance evaluation (Raamana et al 2021). However, while these works aimed at the fully automatic prediction of image quality from unseen scans/sites, we focused on identifying a set of informative QC features as a pre-selection and guide for visual inspection.…”