Model-informed machine learning for multi-component<mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" altimg="si1.svg"><mml:msub><mml:mi>T</mml:mi><mml:mn>2</mml:mn></mml:msub></mml:math>relaxometry

Yu, Thomas; Canales-Rodríguez, Erick Jorge; Pizzolato, Marco; Piredda, Gian Franco; Hilbert, Tom; Fischi-Gómez, Elda; Weigel, Matthias; Baraković, Muhamed; Cuadra, Meritxell Bach; Granziera, Cristina; Kober, Tobias; Thiran, Jean Philippe

doi:10.1016/j.media.2020.101940

Cited by 30 publications

(27 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent works that leverage supervised ML for model parameter estimation in qMRI typically employ one of two training data distributions: (1) parameter combinations obtained from traditional model fitting and the corresponding measured qMRI signals, 4,6,9,11,[14][15][16][17] or (2) parameters sampled uniformly from the entire plausible parameter space with simulated qMRI signals. 5,[18][19][20][21][22][23][24] While (1) uses parameter combinations directly estimated from the data so likely quantifies the model parameters with higher accuracy and precision for a given specific dataset, (2) supports choice of training data distribution, which may help improve generalizability and avoid problems arising from imbalance.…”

Section: Introductionmentioning

confidence: 99%

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Gyori

Palombo

Clark

et al. 2021

Magnetic Resonance in Med

View full text Add to dashboard Cite

Purpose Supervised machine learning (ML) provides a compelling alternative to traditional model fitting for parameter mapping in quantitative MRI. The aim of this work is to demonstrate and quantify the effect of different training data distributions on the accuracy and precision of parameter estimates when supervised ML is used for fitting. Methods We fit a two‐ and three‐compartment biophysical model to diffusion measurements from in‐vivo human brain, as well as simulated diffusion data, using both traditional model fitting and supervised ML. For supervised ML, we train several artificial neural networks, as well as random forest regressors, on different distributions of ground truth parameters. We compare the accuracy and precision of parameter estimates obtained from the different estimation approaches using synthetic test data. Results When the distribution of parameter combinations in the training set matches those observed in healthy human data sets, we observe high precision, but inaccurate estimates for atypical parameter combinations. In contrast, when training data is sampled uniformly from the entire plausible parameter space, estimates tend to be more accurate for atypical parameter combinations but may have lower precision for typical parameter combinations. Conclusion This work highlights that estimation of model parameters using supervised ML depends strongly on the training‐set distribution. We show that high precision obtained using ML may mask strong bias, and visual assessment of the parameter maps is not sufficient for evaluating the quality of the estimates.

show abstract

Section: Introductionmentioning

confidence: 99%

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Gyori

Palombo

Clark

et al. 2021

Magnetic Resonance in Med

View full text Add to dashboard Cite

show abstract

“…Multiple approaches have been proposed to mitigate noise amplification in MWF mapping without further increasing scan time, including the development of temporal and spatial regularization in solving the least‐squares minimization problems 15,16,41–43 and the advent of SNR‐efficient 3D acquisitions such as the 4‐minute FAST‐T2 sequence with very short TE and geometric echo spacing to increase sensitivity to myelin water 9,33 . Recently, several research groups have successfully applied neural networks to the MWF extraction problem with the goal of replicating MWF results of the iterative solvers but with much faster speed 19–21 . Unlike these previous works, which all used MLP network architecture to process the input data voxel by voxel, here we used a 3D UNET CNN architecture with much deeper layers and more weights to judiciously utilize the rich local spatiotemporal pattern inherent in the multi‐echo 3D brain data for whole‐brain processing.…”

Section: Discussionmentioning

confidence: 99%

“…9,33 Recently, several research groups have successfully applied neural networks to the MWF extraction problem with the goal of replicating MWF results of the iterative solvers but with much faster speed. [19][20][21] Unlike these previous works, which all used MLP network architecture to process the input data voxel by voxel, here we used a 3D UNET CNN architecture with much deeper layers and more weights to judiciously utilize the rich local spatiotemporal pattern inherent in the multi-echo 3D brain data for whole-brain processing. The improved ability to aggregate information in time and space across multiple scales of CNN in UNET likely contributes to more effective noise suppression and generally leads to better accuracy for in vivo srNLLS MWF prediction of UNET compared with MLP.…”

Section: Discussionmentioning

confidence: 99%

“…Recently, supervised machine learning based on artificial neural networks has been proposed to provide fast and accurate inference for quantitative MRI analysis 17 . In the context of MWF imaging, previous studies have successfully trained multilayer perceptron (MLP) models 18 to rapidly replicate the output of traditional iterative fitting algorithms for MWF reconstruction 19–21 . In contrast to MLP, in which image voxels are processed independently from each other, deep learning methods based on convolutional neural networks (CNNs) can aggregate signals from all voxels in an image to improve the prediction accuracy.…”

Section: Introductionmentioning

confidence: 99%

“…17 In the context of MWF imaging, previous studies have successfully trained multilayer perceptron (MLP) models 18 to rapidly replicate the output of traditional iterative fitting algorithms for MWF reconstruction. [19][20][21] In contrast to MLP, in which image voxels are processed independently from each other, deep learning methods based on convolutional neural networks (CNNs) can aggregate signals from all voxels in an image to improve the prediction accuracy. In particular, the UNET architecture 22 with hierarchical feature compression, extraction, and concatenation has demonstrated good performance on various image-to-image mapping tasks in MRI, including lesion segmentation, 23,24 image reconstruction, [25][26][27][28] and MR parameter mapping.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Subsecond accurate myelin water fraction reconstruction from FAST‐T₂ data with 3D UNET

Kim

Nguyen

Zhang

et al. 2022

Magnetic Resonance in Med

View full text Add to dashboard Cite

Purpose To develop a 3D UNET convolutional neural network for rapid extraction of myelin water fraction (MWF) maps from six‐echo fast acquisition with spiral trajectory and T2‐prep data and to evaluate its accuracy in comparison with multilayer perceptron (MLP) network. Methods The MWF maps were extracted from 138 patients with multiple sclerosis using an iterative three‐pool nonlinear least‐squares algorithm (NLLS) without and with spatial regularization (srNLLS), which were used as ground‐truth labels to train, validate, and test UNET and MLP networks as a means to accelerate data fitting. Network testing was performed in 63 patients with multiple sclerosis and a numerically simulated brain phantom at SNR of 200, 100 and 50. Results Simulations showed that UNET reduced the MWF mean absolute error by 30.1% to 56.4% and 16.8% to 53.6% over the whole brain and by 41.2% to 54.4% and 21.4% to 49.4% over the lesions for predicting srNLLS and NLLS MWF, respectively, compared to MLP, with better performance at lower SNRs. UNET also outperformed MLP for predicting srNLLS MWF in the in vivo multiple‐sclerosis brain data, reducing mean absolute error over the whole brain by 61.9% and over the lesions by 67.5%. However, MLP yielded 41.1% and 51.7% lower mean absolute error for predicting in vivo NLLS MWF over the whole brain and the lesions, respectively, compared with UNET. The whole‐brain MWF processing time using a GPU was 0.64 seconds for UNET and 0.74 seconds for MLP. Conclusion Subsecond whole‐brain MWF extraction from fast acquisition with spiral trajectory and T2‐prep data using UNET is feasible and provides better accuracy than MLP for predicting MWF output of srNLLS algorithm.

show abstract

Widespread intra‐axonal signal fraction abnormalities in bipolar disorder from multicompartment diffusion MRI: Sensitivity to diagnosis, association with clinical features and pharmacologic treatment

Canales‐Rodríguez,

Verdolini,

Alonso‐Lana

et al. 2023

Human Brain Mapping

Self Cite

View full text Add to dashboard Cite

Despite diffusion tensor imaging (DTI) evidence for widespread fractional anisotropy (FA) reductions in the brain white matter of patients with bipolar disorder, questions remain regarding the specificity and sensitivity of FA abnormalities as opposed to other diffusion metrics in the disorder. We conducted a whole-brain voxel-based Raymond Salvador and Edith Pomarol-Clotet are joint last author (equal contribution, shared senior authorship).

show abstract

Model-informed machine learning for multi-componentT2relaxometry

Cited by 30 publications

References 50 publications

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Subsecond accurate myelin water fraction reconstruction from FAST‐T₂ data with 3D UNET

Widespread intra‐axonal signal fraction abnormalities in bipolar disorder from multicompartment diffusion MRI: Sensitivity to diagnosis, association with clinical features and pharmacologic treatment

Contact Info

Product

Resources

About

Model-informed machine learning for multi-componentT2relaxometry

Cited by 30 publications

References 50 publications

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Training data distribution significantly impacts the estimation of tissue microstructure with machine learning

Subsecond accurate myelin water fraction reconstruction from FAST‐T2 data with 3D UNET

Widespread intra‐axonal signal fraction abnormalities in bipolar disorder from multicompartment diffusion MRI: Sensitivity to diagnosis, association with clinical features and pharmacologic treatment

Contact Info

Product

Resources

About

Subsecond accurate myelin water fraction reconstruction from FAST‐T₂ data with 3D UNET