HYDRA: Hybrid deep magnetic resonance fingerprinting

Song, Pingfan; Eldar, Yonina C.; Mazor, Gal; Rodrigues, Miguel R. D.

doi:10.1002/mp.13727

Cited by 30 publications

(43 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The procedure has the advantage to be semi-parametric but a serious limit is that the components of f are optimized in each dimension separately. As regards application to MRF, the learning strategy has also been proposed by several groups using deep learning tools [11][12][13][14][15][16][17][18]. A major limitation of these methods is that they require a large number of training points to learn many model parameters without overfitting.…”

Section: Proposed Dictionary-based Learning (Dbl) Methodsmentioning

confidence: 99%

“…However, this compression procedure generally decreases parameter accuracy. It has also been proposed to directly find a mapping from the fingerprints to the parameter space using kernel regression [9], maximum likelihood approach [10] or neural network approaches [11][12][13][14][15][16][17][18]. The resulting compact representation offers the advantage over the discrete MRF grid of a continuous exploration of parameter values.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Bayesian Inverse Regression for Vascular Magnetic Resonance Fingerprinting

Boux

Forbes

Arbel

et al. 2021

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Standard parameter estimation from vascular magnetic resonance fingerprinting (MRF) data is based on matching the MRF signals to their best counterparts in a grid of coupled simulated signals and parameters, referred to as a dictionary. To reach a good accuracy, the matching requires an informative dictionary whose cost, in terms of design, storage and exploration, is rapidly prohibitive for even moderate numbers of parameters. In this work, we propose an alternative dictionary-based learning (DBL) approach made of three steps: 1) a quasi-random sampling strategy to produce efficiently an informative dictionary, 2) an inverse statistical regression model to learn from the dictionary a correspondence between fingerprints and parameters, and 3) the use of this mapping to provide both parameter estimates and their confidence indices. Our DBL method is first compared to MRF matching on two types of synthetic signals: scalable and vascular MRF signals. On scalable signals, quasi-random sampling outperforms the grid when using DBL. Dictionaries up to 100 times smaller than in MRF matching, yield a 12 % decreased error on parameter estimates. The confidence indices match the parameter estimation errors (R 2 = 0.95). Then, on vascular signals, dictionary-based methods yield more accurate estimates than the conventional, closed-form expression fitting method with significantly smaller errors on vessel size estimates. On real vascular MRF signals acquired from tumor bearing rats, the DBL method shows less noisy maps than MRF matching. Our DBL proposal effectively reduces the number of simulations required and speeds up parameter estimation, while providing more accurate estimates with their confidence indices.

show abstract

Section: Proposed Dictionary-based Learning (Dbl) Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Bayesian Inverse Regression for Vascular Magnetic Resonance Fingerprinting

Boux

Forbes

Arbel

et al. 2021

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

show abstract

“…Approaches tested include using a Kalman filter, 33 treating quantitative MRI as a nonlinear tomography problem, 30 kernel ridge regression, 32,34 and Deep Learning. 11,[35][36][37][38][39][40][41][42][43][44] Deep Learning in particular has shown promise, reducing errors in relaxometry estimates, 39,40,42 and optimising the dictionary-matching process, 36,41,43 In one study, a four-layer neural network utilising rapid feedforward processing was trained on simulated MRI data and tested on numerical and ISMRM/NIST MRI phantoms. Image reconstruction was accurate and demonstrated image reconstruction up to 5000 times faster, vast storage savings and robustness to noise as compared to conventional MRF dictionary-matching.…”

Section: Non-dictionary Image Reconstruction Methodsmentioning

confidence: 99%

“…Several protocols employ faster scan times, although computing requirements and reconstruction times vary. Approaches tested include using a Kalman filter, 33 treating quantitative MRI as a nonlinear tomography problem, 30 kernel ridge regression, 32,34 and Deep Learning 11,35‐44 …”

Section: Optimising Image Reconstructionmentioning

confidence: 99%

Magnetic resonance fingerprinting: from evolution to clinical applications

Hsieh

Svalbe

2020

J of Medical Radiation Sci

View full text Add to dashboard Cite

In 2013, Magnetic Resonance Fingerprinting (MRF) emerged as a method for fast, quantitative Magnetic Resonance Imaging. This paper reviews the current status of MRF up to early 2020 and aims to highlight the advantages MRF can offer medical imaging professionals. By acquiring scan data as pseudorandom samples, MRF elicits a unique signal evolution, or 'fingerprint', from each tissue type. It matches 'randomised' free induction decay acquisitions against precomputed simulated tissue responses to generate a set of quantitative images of T 1 , T 2 and proton density (PD) with co-registered voxels, rather than as traditional relative T 1-and T 2-weighted images. MRF numeric pixel values retain accuracy and reproducibility between 2% and 8%. MRF acquisition is robust to strong undersampling of k-space. Scan sequences have been optimised to suppress sub-sampling artefacts, while artificial intelligence and machine learning techniques have been employed to increase matching speed and precision. MRF promises improved patient comfort with reduced scan times and fewer image artefacts. Quantitative MRF data could be used to define population-wide numeric biomarkers that classify normal versus diseased tissue. Certification of clinical centres for MRF scan repeatability would permit numeric comparison of sequential images for any individual patient and the pooling of multiple patient images across large, cross-site imaging studies. MRF has to date shown promising results in early clinical trials, demonstrating reliable differentiation between malignant and benign prostate conditions, and normal and sclerotic hippocampal tissue. MRF is now undergoing small-scale trials at several sites across the world; moving it closer to routine clinical application.

show abstract

“…To achieve computer‐aided diagnosis, deep learning has been widely applied in medical imaging over the past years. Successful cases of applying deep learning include shear wave elastography, computed tomography, optical coherence tomography, magnetic resonance imaging, and transrectal ultrasound imaging . Deep learning has also recently been used to differentiate benign and malignant breast tumors in ultrasound images.…”

Section: Introductionmentioning

confidence: 99%

Breast tumor classification through learning from noisy labeled ultrasound images

et al. 2019

View full text Add to dashboard Cite

Purpose To train deep learning models to differentiate benign and malignant breast tumors in ultrasound images, we need to collect many training samples with clear labels. In general, biopsy results can be used as benign/malignant labels. However, most clinical samples generally do not have biopsy results. Previous works have proposed generating benign/malignant labels according to Breast Imaging, Reporting and Data System (BI‐RADS) ratings. However, this approach will cause noisy labels, which means that the benign/malignant labels produced from BI‐RADS diagnoses may be inconsistent with the ground truths. Consequently, deep models will overfit the noisy labels and hence obtain poor generalization performance. In this work, we mainly focus on how to reduce the negative effect of noisy labels when they are used to train breast tumor classification models. Methods We propose an effective approach called noise filter network (NF‐Net) to address the problem of noisy labels when training breast tumor classification models. Specifically, to prevent deep models from overfitting the noisy labels, we propose incorporating two softmax layers for classification. Additionally, to strengthen the effect of clean labels, we design a teacher–student module for distilling the knowledge of clean labels. Results We conduct extensive comparisons with the existing works on addressing noisy labels. Our method achieves a classification accuracy of 73%, with a precision of 69%, recall of 80%, and F1‐score of 0.74. This result is significantly better than those of the existing state‐of‐the‐art works on addressing noisy labels. Conclusions This work provides a means to overcome the label shortage problem in training breast tumor classification models. Specifically, we can generate benign/malignant labels according to the BI‐RADS ratings. Although this approach will cause noisy labels, the design of NF‐Net can effectively reduce the negative effect of such labels.

show abstract

HYDRA: Hybrid deep magnetic resonance fingerprinting

Cited by 30 publications

References 40 publications

Bayesian Inverse Regression for Vascular Magnetic Resonance Fingerprinting

Bayesian Inverse Regression for Vascular Magnetic Resonance Fingerprinting

Magnetic resonance fingerprinting: from evolution to clinical applications

Breast tumor classification through learning from noisy labeled ultrasound images

Contact Info

Product

Resources

About