Accurate anatomical head segmentations: a data set for biomedical simulations

Farcito, Silvia; Puonti, Oula; Montanaro, Hazael; Saturnino, Guilherme B.; Nielsen, Jakob Skov; Madsen, Camilla Gøbel; Siebner, Hartwig R.; Neufeld, Esra; Kuster, Niels; Lloyd, Bryn; Thielscher, Axel

doi:10.1109/embc.2019.8857041

Cited by 8 publications

(13 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By The added extra-cerebral labels (left) are obtained with a Bayesian segmentation approach [106], for which we only had access to the atlas (not to the manual delineations used for its construction [107]).…”

Section: Discussionmentioning

confidence: 99%

“…To further increase the variability of the training label maps, we apply the following steps. First, we complete them with extra-cerebral labels obtained with a Bayesian segmentation approach [106], for which we only had access to the atlas (not to the delineations used for its construction [107]). Then, during training, one of the following options is randomly selected at each mini-batch: (i) keep all labels, (ii) suppress all extra-cerebral labels, or (iii) suppress all extra-cerebral labels except the CSF (Supplement 2).…”

Section: Training Segmentations and Population Robustnessmentioning

confidence: 99%

“…Using these different versions builds robustness to white matter lesions and to (possibly imperfect) skull stripping. The added extra-cerebral labels (left) are obtained with a Bayesian segmentation approach[106], for which we only had access to the atlas (not to the manual delineations used for its construction[107]).…”

mentioning

confidence: 99%

See 2 more Smart Citations

SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

Billot¹,

Greve²,

Puonti³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Despite advances in data augmentation and transfer learning, convolutional neural networks (CNNs) have difficulties generalising to unseen target domains. When applied to segmentation of brain MRI scans, CNNs are highly sensitive to changes in resolution and contrast: even within the same MR modality, decreases in performance can be observed across datasets. We introduce SynthSeg, the first segmentation CNN agnostic to brain MRI scans of any contrast and resolution. SynthSeg is trained with synthetic data sampled from a generative model inspired by Bayesian segmentation. Crucially, we adopt a domain randomisation strategy where we fully randomise the generation parameters to maximise the variability of the training data. Consequently, SynthSeg can segment preprocessed and unpreprocessed real scans of any target domain, without retraining or fine-tuning. Because SynthSeg only requires segmentations to be trained (no images), it can learn from label maps obtained automatically from existing datasets of different populations (e.g., with atrophy and lesions), thus achieving robustness to a wide range of morphological variability. We demonstrate SynthSeg on 5,500 scans of 6 modalities and 10 resolutions, where it exhibits unparalleled generalisation compared to supervised CNNs, test time adaptation, and Bayesian segmentation. The code and trained model are available at github.com/BBillot/SynthSeg.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Training Segmentations and Population Robustnessmentioning

confidence: 99%

See 1 more Smart Citation

SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

Billot¹,

Greve²,

Puonti³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The dataset used in this study consists of 20 subjects each of which had a T1- and a T2-weighted MRI scan (for details on this dataset, please see Farcito et al, 2019). Importantly, manual segmentations of 16 tissue classes were available for all subjects.…”

Section: Methodsmentioning

confidence: 99%

Evaluating the Influence of Anatomical Accuracy and Electrode Positions on EEG Forward Solutions

Nielsen

Puonti

Xue

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Generating realistic volume conductor models for forward calculations in electroencephalography is not trivial and several factors contribute to the accuracy of such models, two of which are its anatomical accuracy and the accuracy with which electrode positions are known. Here, we investigate effects of anatomical accuracy by comparing forward solutions from SimNIBS, a tool which allows state-of-the-art anatomical modeling, with well-established pipelines in MNE-Python and FieldTrip. We also compare different ways of specifying electrode locations when digitized positions are not available such as transformation of measured positions from standard space and transformation of a manufacturer layout. Substantial effects of anatomical accuracy were seen throughout the entire brain both in terms of field topography and magnitude with SimNIBS generally being more accurate than the pipelines in MNE-Python and FieldTrip. Topographic and magnitude effects were particularly pronounced for MNE-Python which uses a three-layer boundary element model. We attribute these mainly to differences in skull and cerebrospinal fluid modeling. Effects of electrode specification method were evident in occipital and posterior areas when using a transformed manufacturer layout whereas transforming measured positions from standard space generally resulted in smaller errors. We suggest modeling the anatomy of the volume conductor as accurately possible and we hope to facilitate this by making it easy to export simulations from SimNIBS to MNE-Python and FieldTrip for further analysis. Likewise, if digitized electrode positions are not available, a set of measured positions on a standard head template may be preferable to those specified by the manufacturer.

show abstract

“…Simulations involving three different human head models (S1, S2, S3) from [37] were performed using the P 1 and the S eff transducer model (see Figure 8). These head models were created from multi-modal image data (different MRI sequences, as well as CT), permitting the consideration of skull heterogeneity.…”

Section: Human Head Modelsmentioning

confidence: 99%

Transducer modeling for accurate acoustic simulations of transcranial focused ultrasound stimulation

Pasquinelli¹,

Montanaro²,

Lee³

et al. 2020

J. Neural Eng.

Self Cite

View full text Add to dashboard Cite

Objective: Low-intensity transcranial ultrasound stimulation (TUS) is emerging as noninvasive brain stimulation technique with superior spatial resolution and the ability to reach deep brain areas. Medical image-based computational modeling could be an important tool for individualized TUS dose control and targeting optimization, but requires further validation. This study aims to assess the impact of the transducer model on the accuracy of the simulations.Approach: Using hydrophone measurements, the acoustic beam of a single-element focused transducer (SEFT) with a flat piezoelectric disc and an acoustic lens was characterized. The acoustic beam was assessed in a homogeneous water bath and after transmission through obstacles (3D-printed shapes and skull samples). The acoustic simulations employed the finite-difference time-domain method and were informed by computed tomography (CT) images of the obstacles. Transducer models of varying complexity were tested, representing the SEFT either as a surface boundary condition with variable curvature, or also accounting for its internal geometry. In addition, a back-propagated pressure distribution from the first measurement plane was used as source model. The simulations and measurements were quantitatively compared using key metrics for peak location, focus size, intensity and spatial distribution.Main results: While a surface boundary with an adapted, 'effective' curvature radius based on the specifications given by the manufacturer could reproduce the measured focus location and size in a homogeneous water bath, it regularly failed to accurately predict the beam after obstacle transmission. In contrast, models that were based on a one-time 1 calibration to the homogeneous water bath measurements performed substantially better in all cases with obstacles. For one of the 3D-printed obstacles, the simulated intensities deviated substantially from the measured ones, irrespective of the transducer model. We attribute this finding to a standing wave effect, and further studies should clarify its relevance for accurate simulations of skull transmission.Significance: Validated transducer models are important to ensure accurate simulations of the acoustic beam of SEFTs, in particular in the presence of obstacles such as the skull.

show abstract

Accurate anatomical head segmentations: a data set for biomedical simulations

Cited by 8 publications

References 25 publications

SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining

Evaluating the Influence of Anatomical Accuracy and Electrode Positions on EEG Forward Solutions

Transducer modeling for accurate acoustic simulations of transcranial focused ultrasound stimulation

Contact Info

Product

Resources

About