A deep learning method for image-based subject-specific local SAR assessment

E.F. Meliadò (Corresponding author), A.J.E. Raaijmakers, A. Sbrizzi, B.R. Steensma, M. Maspero, M.H.F. Savenije, P.R. Luijten, C.A.T. van den Berg

Onderzoeksoutput: Bijdrage aan tijdschriftTijdschriftartikelAcademicpeer review

7 Downloads (Pure)

Uittreksel

Purpose: Local specific absorption rate (SAR) cannot be measured and is usually evaluated by offline numerical simulations using generic body models that of course will differ from the patient's anatomy. An additional safety margin is needed to include this intersubject variability. In this work, we present a deep learning–based method for image-based subject-specific local SAR assessment. We propose to train a convolutional neural network to learn a “surrogate SAR model” to map the relation between subject-specific (Formula presented.) maps and the corresponding local SAR. Method: Our database of 23 subject-specific models with an 8–transmit channel body array for prostate imaging at 7 T was used to build 5750 training samples. These synthetic complex (Formula presented.) maps and local SAR distributions were used to train a conditional generative adversarial network. Extra penalization for local SAR underestimation errors was included in the loss function. In silico and in vivo validation were performed. Results: In silico cross-validation shows a good qualitative and quantitative match between predicted and ground-truth local SAR distributions. The peak local SAR estimation error distribution shows a mean overestimation error of 15% with 13% probability of underestimation. The higher accuracy of the proposed method allows the use of less conservative safety factors compared with standard procedures. In vivo validation shows that the method is applicable with realistic measurement data with impressively good qualitative and quantitative agreement to simulations. Conclusion: The proposed deep learning method allows online image-based subject-specific local SAR assessment. It greatly reduces the uncertainty in current state-of-the-art SAR assessment methods, reducing the time in the examination protocol by almost 25%.

Originele taal-2Engels
Pagina's (van-tot)695-711
Aantal pagina's17
TijdschriftMagnetic Resonance in Medicine
Volume83
Nummer van het tijdschrift2
DOI's
StatusGepubliceerd - 1 feb 2020

Vingerafdruk

Learning
Computer Simulation
Safety
Uncertainty
Prostate
Anatomy
Databases

Citeer dit

Meliadò, E. F., Raaijmakers, A. J. E., Sbrizzi, A., Steensma, B. R., Maspero, M., Savenije, M. H. F., ... van den Berg, C. A. T. (2020). A deep learning method for image-based subject-specific local SAR assessment. Magnetic Resonance in Medicine, 83(2), 695-711. https://doi.org/10.1002/mrm.27948
Meliadò, E.F. ; Raaijmakers, A.J.E. ; Sbrizzi, A. ; Steensma, B.R. ; Maspero, M. ; Savenije, M.H.F. ; Luijten, P.R. ; van den Berg, C.A.T. / A deep learning method for image-based subject-specific local SAR assessment. In: Magnetic Resonance in Medicine. 2020 ; Vol. 83, Nr. 2. blz. 695-711.
@article{d5909ad89d3a41dc96cb5e153cbbe188,
title = "A deep learning method for image-based subject-specific local SAR assessment",
abstract = "Purpose: Local specific absorption rate (SAR) cannot be measured and is usually evaluated by offline numerical simulations using generic body models that of course will differ from the patient's anatomy. An additional safety margin is needed to include this intersubject variability. In this work, we present a deep learning–based method for image-based subject-specific local SAR assessment. We propose to train a convolutional neural network to learn a “surrogate SAR model” to map the relation between subject-specific (Formula presented.) maps and the corresponding local SAR. Method: Our database of 23 subject-specific models with an 8–transmit channel body array for prostate imaging at 7 T was used to build 5750 training samples. These synthetic complex (Formula presented.) maps and local SAR distributions were used to train a conditional generative adversarial network. Extra penalization for local SAR underestimation errors was included in the loss function. In silico and in vivo validation were performed. Results: In silico cross-validation shows a good qualitative and quantitative match between predicted and ground-truth local SAR distributions. The peak local SAR estimation error distribution shows a mean overestimation error of 15{\%} with 13{\%} probability of underestimation. The higher accuracy of the proposed method allows the use of less conservative safety factors compared with standard procedures. In vivo validation shows that the method is applicable with realistic measurement data with impressively good qualitative and quantitative agreement to simulations. Conclusion: The proposed deep learning method allows online image-based subject-specific local SAR assessment. It greatly reduces the uncertainty in current state-of-the-art SAR assessment methods, reducing the time in the examination protocol by almost 25{\%}.",
keywords = "convolutional neural network, deep learning, parallel transmit, specific absorption rate, subject-specific SAR assessment, ultrahigh-field MRI",
author = "E.F. Meliad{\`o} and A.J.E. Raaijmakers and A. Sbrizzi and B.R. Steensma and M. Maspero and M.H.F. Savenije and P.R. Luijten and {van den Berg}, C.A.T.",
year = "2020",
month = "2",
day = "1",
doi = "10.1002/mrm.27948",
language = "English",
volume = "83",
pages = "695--711",
journal = "Magnetic Resonance in Medicine",
issn = "0740-3194",
publisher = "Wiley",
number = "2",

}

Meliadò, EF, Raaijmakers, AJE, Sbrizzi, A, Steensma, BR, Maspero, M, Savenije, MHF, Luijten, PR & van den Berg, CAT 2020, 'A deep learning method for image-based subject-specific local SAR assessment', Magnetic Resonance in Medicine, vol. 83, nr. 2, blz. 695-711. https://doi.org/10.1002/mrm.27948

A deep learning method for image-based subject-specific local SAR assessment. / Meliadò, E.F. (Corresponding author); Raaijmakers, A.J.E.; Sbrizzi, A.; Steensma, B.R.; Maspero, M.; Savenije, M.H.F.; Luijten, P.R.; van den Berg, C.A.T.

In: Magnetic Resonance in Medicine, Vol. 83, Nr. 2, 01.02.2020, blz. 695-711.

Onderzoeksoutput: Bijdrage aan tijdschriftTijdschriftartikelAcademicpeer review

TY - JOUR

T1 - A deep learning method for image-based subject-specific local SAR assessment

AU - Meliadò, E.F.

AU - Raaijmakers, A.J.E.

AU - Sbrizzi, A.

AU - Steensma, B.R.

AU - Maspero, M.

AU - Savenije, M.H.F.

AU - Luijten, P.R.

AU - van den Berg, C.A.T.

PY - 2020/2/1

Y1 - 2020/2/1

N2 - Purpose: Local specific absorption rate (SAR) cannot be measured and is usually evaluated by offline numerical simulations using generic body models that of course will differ from the patient's anatomy. An additional safety margin is needed to include this intersubject variability. In this work, we present a deep learning–based method for image-based subject-specific local SAR assessment. We propose to train a convolutional neural network to learn a “surrogate SAR model” to map the relation between subject-specific (Formula presented.) maps and the corresponding local SAR. Method: Our database of 23 subject-specific models with an 8–transmit channel body array for prostate imaging at 7 T was used to build 5750 training samples. These synthetic complex (Formula presented.) maps and local SAR distributions were used to train a conditional generative adversarial network. Extra penalization for local SAR underestimation errors was included in the loss function. In silico and in vivo validation were performed. Results: In silico cross-validation shows a good qualitative and quantitative match between predicted and ground-truth local SAR distributions. The peak local SAR estimation error distribution shows a mean overestimation error of 15% with 13% probability of underestimation. The higher accuracy of the proposed method allows the use of less conservative safety factors compared with standard procedures. In vivo validation shows that the method is applicable with realistic measurement data with impressively good qualitative and quantitative agreement to simulations. Conclusion: The proposed deep learning method allows online image-based subject-specific local SAR assessment. It greatly reduces the uncertainty in current state-of-the-art SAR assessment methods, reducing the time in the examination protocol by almost 25%.

AB - Purpose: Local specific absorption rate (SAR) cannot be measured and is usually evaluated by offline numerical simulations using generic body models that of course will differ from the patient's anatomy. An additional safety margin is needed to include this intersubject variability. In this work, we present a deep learning–based method for image-based subject-specific local SAR assessment. We propose to train a convolutional neural network to learn a “surrogate SAR model” to map the relation between subject-specific (Formula presented.) maps and the corresponding local SAR. Method: Our database of 23 subject-specific models with an 8–transmit channel body array for prostate imaging at 7 T was used to build 5750 training samples. These synthetic complex (Formula presented.) maps and local SAR distributions were used to train a conditional generative adversarial network. Extra penalization for local SAR underestimation errors was included in the loss function. In silico and in vivo validation were performed. Results: In silico cross-validation shows a good qualitative and quantitative match between predicted and ground-truth local SAR distributions. The peak local SAR estimation error distribution shows a mean overestimation error of 15% with 13% probability of underestimation. The higher accuracy of the proposed method allows the use of less conservative safety factors compared with standard procedures. In vivo validation shows that the method is applicable with realistic measurement data with impressively good qualitative and quantitative agreement to simulations. Conclusion: The proposed deep learning method allows online image-based subject-specific local SAR assessment. It greatly reduces the uncertainty in current state-of-the-art SAR assessment methods, reducing the time in the examination protocol by almost 25%.

KW - convolutional neural network

KW - deep learning

KW - parallel transmit

KW - specific absorption rate

KW - subject-specific SAR assessment

KW - ultrahigh-field MRI

UR - http://www.scopus.com/inward/record.url?scp=85071754700&partnerID=8YFLogxK

U2 - 10.1002/mrm.27948

DO - 10.1002/mrm.27948

M3 - Article

C2 - 31483521

AN - SCOPUS:85071754700

VL - 83

SP - 695

EP - 711

JO - Magnetic Resonance in Medicine

JF - Magnetic Resonance in Medicine

SN - 0740-3194

IS - 2

ER -