2020
DOI: 10.1016/j.neuroimage.2020.117012
|View full text |Cite
|
Sign up to set email alerts
|

FastSurfer - A fast and accurate deep learning based neuroimaging pipeline

Abstract: Traditional neuroimage analysis pipelines involve computationally intensive, time-consuming optimization steps, and thus, do not scale well to large cohort studies with thousands or tens of thousands of individuals. In this work we propose a fast and accurate deep learning based neuroimaging pipeline for the automated processing of structural human brain MRI scans, replicating FreeSurfer’s anatomical segmentation including surface reconstruction and cortical parcellation. To this end, we introduce an advanced … Show more

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1

Citation Types

9
309
0

Year Published

2020
2020
2024
2024

Publication Types

Select...
6
3
1

Relationship

1
9

Authors

Journals

citations
Cited by 361 publications
(318 citation statements)
references
References 85 publications
9
309
0
Order By: Relevance
“…A potential limitation of our work is that the framework is not generative and therefore cannot be easily used in conjunction with tools such as VBM or Freesurfer, although could be added into deep learning equivalents such as Fastsurfer [23], especially as more tools gain CNN-based equivalents. The decision to not make the framework generative was due to generative CNN methods needing large amounts of data to train or, even in some examples, paired data, meaning that they are not applicable to most real life neuroimage studies, whereas we have shown that our framework is applicable even on highly multi-site datasets with limited examples from each site, such as the ABIDE data.…”
Section: Discussionmentioning
confidence: 99%
“…A potential limitation of our work is that the framework is not generative and therefore cannot be easily used in conjunction with tools such as VBM or Freesurfer, although could be added into deep learning equivalents such as Fastsurfer [23], especially as more tools gain CNN-based equivalents. The decision to not make the framework generative was due to generative CNN methods needing large amounts of data to train or, even in some examples, paired data, meaning that they are not applicable to most real life neuroimage studies, whereas we have shown that our framework is applicable even on highly multi-site datasets with limited examples from each site, such as the ABIDE data.…”
Section: Discussionmentioning
confidence: 99%
“…As such, we removed two outliers as determined by the differences in white surface reconstructions between the test and retest scans for our reproducibility analyses. Future work for improving SBCI should also focus on more robust surface reconstruction e.g., better segmentation and surface reconstruction methods [Zhao et al, 2019;Henschel et al, 2020], or collecting and incorporating high resolution T2w or FLAIR images to the T1w processing pipeline [Van Essen et al, 2013;Glasser et al, 2013;Renvall et al, 2016;Zaretskaya et al, 2018].…”
Section: Discussionmentioning
confidence: 99%
“…Restricting the scope to 3T data only, recent DL-based methods such as QuickNat (Roy et al, 2019), MeshNet (Fedorov et al, 2017;McClure et al, 2019), NeuroNet (Rajchl et al, 2018), DeepNAT (Wachinger et al, 2018), and FastSurfer (Henschel et al, 2020) have been the most effective solutions among those which proposed to obtain a whole brain segmentation. However, a common trait of all the aforementioned methods is that none of them fully exploit the 3D spatial nature of MRI data, thus making segmentation accuracy sub-optimal.…”
Section: Deep Learning Methods For Mri Segmentationmentioning
confidence: 99%