OpenLB
Open Library of Bioscience
Advanced Search
Medical physics
Mar 16, 2025;

Quantitative susceptibility mapping via deep neural networks with iterative reverse concatenations and recurrent modules.

Min Li, Chen Chen, Zhuang Xiong, Yin Liu, Pengfei Rong, Shanshan Shan, Feng Liu, Hongfu Sun, Yang Gao
Author Information
  1. Min Li: School of Computer Science and Engineering, Central South University, Changsha, China.
  2. Chen Chen: School of Computer Science and Engineering, Central South University, Changsha, China.
  3. Zhuang Xiong: School of Electrical Engineering and Computer Science, University of Queensland, Brisbane, Australia.
  4. Yin Liu: Department of Radiology, The Third Xiangya Hospital, Central South University, Changsha, China.
  5. Pengfei Rong: Department of Radiology, The Third Xiangya Hospital, Central South University, Changsha, China.
  6. Shanshan Shan: State Key Laboratory of Radiation, Medicine and Protection, School for Radiological and Interdisciplinary Sciences (RAD-X), Collaborative Innovation Center of Radiation Medicine of Jiangsu Higher Education Institutions, Soochow University, Suzhou, China.
  7. Feng Liu: School of Electrical Engineering and Computer Science, University of Queensland, Brisbane, Australia.
  8. Hongfu Sun: School of Engineering, University of Newcastle, Newcastle, Australia.
  9. Yang Gao: School of Computer Science and Engineering, Central South University, Changsha, China.
PMID: 40089979 DOI: 10.1002/mp.17747

Abstract

BACKGROUND: Quantitative susceptibility mapping (QSM) is a post-processing magnetic resonance imaging (MRI) technique that extracts the distribution of tissue susceptibilities and holds significant promise in the study of neurological diseases. However, the ill-conditioned nature of dipole inversion often results in noise and artifacts during QSM reconstruction from the tissue field. Deep learning methods have shown great potential in addressing these issues; however, most existing approaches rely on basic U-net structures, leading to limited performances and reconstruction artifacts sometimes.
PURPOSE: This study aims to develop a novel deep learning-based method, IRQSM, for improving QSM reconstruction accuracy while mitigating noise and artifacts by leveraging a unique network architecture that enhances latent feature utilization.
METHODS: IRQSM, an advanced U-net architecture featuring four iterations of reverse concatenations and middle recurrent modules, was proposed to optimize feature fusion and improve QSM accuracy, and comparative experiments based on both simulated and in vivo datasets were carried out to compare IRQSM with two traditional iterative methods (iLSQR, MEDI) and four recently proposed deep learning methods (U-net, xQSM, LPCNN, and MoDL-QSM).
RESULTS: In this work, IRQSM outperformed all other methods in reducing artifacts and noise in QSM images. It achieved on average the lowest XSIM (84.81%) in simulations, showing improvements of 12.80%, 12.68%, 18.66%, 10.49%, 25.57%, and 19.78% over iLSQR, MEDI, U-net, xQSM, LPCNN, and MoDL-QSM, respectively, and yielded results with the least artifacts on the in vivo data and present the most visually appealing results. In the meantime, it successfully alleviated the over-smoothing and susceptibility underestimation in LPCNN results.
CONCLUSION: Overall, the proposed IRQSM showed superior QSM results compared to iterative and deep learning-based methods, offering a more accurate QSM solution for clinical applications.

Keywords

IR2QSM deep learning dipole inversion quantitative susceptibility mapping recurrent module reverse concatenation

References

  1. De Rochefort L, Brown R, Prince MR, Wang Y. Quantitative MR susceptibility mapping using piece���wise constant regularized inversion of the magnetic field. Magn Reson Med. 2008;60(4):1003���1009.
  2. Langkammer C, Schweser F, Krebs N, et al. Quantitative susceptibility mapping (QSM) as a means to measure brain iron? A post mortem validation study. Neuroimage. 2012;62(3):1593���1599.
  3. Liu C, Wei H, Gong NJ, et al. Quantitative susceptibility mapping: contrast mechanisms and clinical applications. Tomography. 2015;1(1):3���17.
  4. Wang Y, Liu T. Quantitative susceptibility mapping (QSM): decoding MRI data for a tissue magnetic biomarker. Magn Reson Med. 2015;73(1):82���101.
  5. Acosta���Cabronero J, Cardenas���Blanco A, Betts MJ, et al. The whole���brain pattern of magnetic susceptibility perturbations in Parkinson's disease. Brain. 2017;140(1):118���131.
  6. Acosta���Cabronero J, Williams GB, Cardenas���Blanco A, Arnold RJ, Lupson V, Nestor PJ. In vivo quantitative susceptibility mapping (QSM) in Alzheimer's disease. PLoS One. 2013;8(11):e81093.
  7. Ayton S, Fazlollahi A, Bourgeat P, et al. Cerebral quantitative susceptibility mapping predicts amyloid���beta���related cognitive decline. Brain. 2017;140(8):2112���2119.
  8. Deistung A, Schweser F, Reichenbach JR. Overview of quantitative susceptibility mapping. NMR Biomed. 2017;30(4).
  9. Elkady AM, Cobzas D, Sun H, Blevins G, Wilman AH. Discriminative analysis of regional evolution of iron and myelin/calcium in deep gray matter of multiple sclerosis and healthy subjects. J Magn Reson Imaging. 2018:652���668.
  10. Langkammer C, Pirpamer L, Seiler S, et al. Quantitative susceptibility mapping in Parkinson's disease. PLoS One. 2016;11(9):e0162460.
  11. Sun H, Walsh AJ, Lebel RM, et al. Validation of quantitative susceptibility mapping with Perls' iron staining for subcortical gray matter. Neuroimage. 2015;105:486���492.
  12. Juh��s M, Sun H, Brown MRG, et al. Deep grey matter iron accumulation in alcohol use disorder. NeuroImage. 2017;148:115���122.
  13. Bilgic B, Pfefferbaum A, Rohlfing T, Sullivan EV, Adalsteinsson E. MRI estimates of brain iron concentration in normal aging using quantitative susceptibility mapping. Neuroimage. 2012;59(3):2625���2635.
  14. Chen L, Soldan A, Oishi K, et al. Quantitative susceptibility mapping of brain iron and �����amyloid in MRI and PET relating to cognitive performance in cognitively normal older adults. Radiology. 2021;298(2):353���362.
  15. Treit S, Naji N, Seres P, et al. R2* and quantitative susceptibility mapping in deep gray matter of 498 healthy controls from 5 to 90 years. Human Brain Mapping. 2021;42(14):4597���4610.
  16. Kunieda T, Togo R, Nishioka N, et al. Estimation of amyloid����� positivity using QSM images considering age information. In 2023 International Conference on Consumer Electronics���Taiwan (ICCE���Taiwan) (pp. 165���166). IEEE. https://doi.org/10.1109/ICCE���Taiwan58799.2023.10226714
  17. Poynton CB, Jenkinson M, Adalsteinsson E, Sullivan EV, Pfefferbaum A, Wells Iii W. Quantitative susceptibility mapping by inversion of a perturbation field model: correlation with brain iron in normal aging. IEEE Trans Med Imaging. 2015;34(1):339���353.
  18. De A, Sun H, Emery DJ, Butcher KS, Wilman AH. Rapid quantitative susceptibility mapping of intracerebral hemorrhage. J Magn Reson Imaging. 2020;51(3):712���718.
  19. Sun H, Klahr AC, Kate M, et al. Quantitative susceptibility mapping for following intracranial hemorrhage. Radiology. 2018;288(3):830���839.
  20. Wang S, Lou M, Liu T, Cui D, Chen X, Wang Y. Hematoma volume measurement in gradient echo MRI using quantitative susceptibility mapping. Stroke. 2013;44(8):2315���2317.
  21. Balla DZ, Sanchez���Panchuelo RM, Wharton SJ, et al. Functional quantitative susceptibility mapping (fQSM). Neuroimage. 2014;100:112���124.
  22. Ma Y, Sun H, Cho J, et al. Cerebral OEF quantification: a comparison study between quantitative susceptibility mapping and dual���gas calibrated BOLD imaging. Magn Reson Med. 2020;83(1):68���82.
  23. Sun H, Seres P, Wilman A. Structural and functional quantitative susceptibility mapping from standard fMRI studies. NMR Biomed. 2017;30(4):e3619.
  24. Liu T, Spincemaille P, De Rochefort L, Kressler B, Wang Y. Calculation of susceptibility through multiple orientation sampling (COSMOS): a method for conditioning the inverse problem from measured magnetic field map to susceptibility source image in MRI. Magn Reson Med. 2008;61(1):196���204.
  25. Shmueli K, De Zwart JA, Van Gelderen P, Li TQ, Dodd SJ, Duyn JH. Magnetic susceptibility mapping of brain tissue in vivo using MRI phase data. Magn Reson Med. 2009;62(6):1510���1522.
  26. Li W, Wang N, Yu F, et al. A method for estimating and removing streaking artifacts in quantitative susceptibility mapping. Neuroimage. 2015;108:111���122.
  27. Liu J, Liu T, De Rochefort L, et al. Morphology enabled dipole inversion for quantitative susceptibility mapping using structural consistency between the magnitude image and the susceptibility map. Neuroimage. 2012;59(3):2560���2568.
  28. Liu T, Xu W, Spincemaille P, Avestimehr AS, Wang Y. Accuracy of the morphology enabled dipole inversion (MEDI) algorithm for quantitative susceptibility mapping in MRI. IEEE Trans Med Imaging. 2012;31(3):816���824.
  29. Bao L, Li X, Cai C, Chen Z, Van Zijl PCM. Quantitative susceptibility mapping using structural feature based collaborative reconstruction(SFCR) in the human brain. IEEE Trans Med Imaging. 2016;35(9):2040���2050.
  30. Wei H, Dibb R, Zhou Y, et al. Streaking artifact reduction for quantitative susceptibility mapping of sources with large dynamic range. NMR Biomed. 2015;28(10):1294���1303.
  31. Sun H, Ma Y, MacDonald ME, Pike GB. Whole head quantitative susceptibility mapping using a least���norm direct dipole inversion method. Neuroimage. 2018;179:166���175.
  32. Yoon J, Gong E, Chatnuntawech I, et al. Quantitative susceptibility mapping using deep neural network: qSMnet. Neuroimage. 2018;179:199���206.
  33. Jung W, Yoon J, Ji S, et al. Exploring linearity of deep neural network trained QSM: qSMnet+. Neuroimage. 2020;211:116619.
  34. Chen Y, Jakary A, Avadiappan S, Hess CP, Lupo JM. QSMGAN: improved quantitative susceptibility mapping using 3D Generative adversarial Networks with increased receptive field. Neuroimage. 2020;207:116389.
  35. Lai K���W, Aggarwal M, van Zijl P, Li X, Sulam J. Learned Proximal Networks for Quantitative Susceptibility Mapping. In Medical Image Computing and Computer Assisted Intervention���MICCAI 2020: 23rd International Conference, 2020: 125���135.
  36. Lu Z, Li J, Wang C, et al. S2Q���Net: mining the high���pass filtered phase data in susceptibility weighted imaging for quantitative susceptibility mapping. IEEE J Biomed Health Inform. 2022;26(8):3938���3949.
  37. Feng R, Zhao J, Wang He, et al. MoDL���QSM: model���based deep learning for quantitative susceptibility mapping. Neuroimage. 2021;240:118376.
  38. Bollmann S, Rasmussen KGB, Kristensen M, et al. DeepQSM���using deep learning to solve the dipole inversion for quantitative susceptibility mapping. Neuroimage. 2019;195:373���383.
  39. Gao Y, Zhu X, Moffat BA, et al. xQSM: quantitative susceptibility mapping with octave convolutional and noise���regularized neural networks. NMR Biomed. 2021;34(3):e4461.
  40. Bao L, Zhang H, Liao Z. A spatially adaptive regularization based three���dimensional reconstruction network for quantitative susceptibility mapping. Phys Med Biol. 2024;69(4):045030.
  41. Xiong Z, Gao Y, Liu F, Sun H. Affine transformation edited and refined deep neural network for quantitative susceptibility mapping. Neuroimage. 2023;267:119842.
  42. Xiong Z, Gao Y, Liu Y, et al. Quantitative susceptibility mapping through model���based deep image prior (MoDIP). NeuroImage. 2024;291:120583.
  43. Oh G, Bae H, Ahn HS, Park SH, Moon WJ, Ye JC. Unsupervised resolution���agnostic quantitative susceptibility mapping using adaptive instance normalization. Med Image Anal. 2022;79:102477.
  44. Wei H, Cao S, Zhang Y, et al. Learning���based single���step quantitative susceptibility mapping reconstruction without brain extraction. Neuroimage. 2019;202:116064.
  45. Gao Y, Xiong Z, Fazlollahi A, et al. Instant tissue field and magnetic susceptibility mapping from MRI raw phase using Laplacian enhanced deep neural networks. Neuroimage. 2022;259:119410.
  46. Gao Y, Xiong Z, Shan S, et al. Plug���and���Play latent feature editing for orientation���adaptive quantitative susceptibility mapping neural networks. Med Image Anal. 2024;94:103160.
  47. Oktay O, Schlemper J, Le Folgoc L, et al. Attention u���net: Learning where to look for the pancreas. In Medical Imaging with Deep Learning. https://openreview.net/forum?id=Skft7cijM
  48. Zhou Z, Rahman Siddiquee MM, Tajbakhsh N, Liang J. Unet++: a nested u���net architecture for medical image segmentation. in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support: 4th International Workshop, DLMIA 2018, and 8th International Workshop, ML���CDS 2018, Held in Conjunction with MICCAI 2018, Granada, Spain, September 20, xs2018, Proceedings 4, 2018:3���11: Springer.
  49. Elman JL. Finding structure in time. Cogn Sci. 2010;14(2):179���211.
  50. Hochreiter S, Schmidhuber J. Long short���term memory. Neural Comput. 1997;9(8):1735���1780.
  51. Koch KM, Papademetris X, Rothman DL, Graaf RADe. Rapid calculations of susceptibility���induced magnetostatic field perturbations for in vivo magnetic resonance. Phys Med Biol. 2006;51(24):6381���6402.
  52. Marques JP, Bowtell R. Application of a Fourier���based method for rapid calculation of field inhomogeneity due to spatial variation of magnetic susceptibility. Concepts Magn Reson B. 2005;25B(1):65���78.
  53. Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. Dropout: a simple way to prevent neural networks from overfitting. J Machine Learn Res. 2014;15(1):1929���1958.
  54. Alom MZ, Hasan M, Yakopcic C, Taha TM, Asari VK. Recurrent residual convolutional neural network based on u���net (r2u���net) for medical image segmentation. arXiv preprint arXiv. 2018:06955.
  55. Wang W, Yu K, Hugonot J, Fua P, Salzmann M. Recurrent U���Net for resource���constrained segmentation. In Proceedings of the IEEE/CVF international conference on computer vision, 2019: 2142���2151.
  56. Sherstinsky A. Fundamentals of recurrent neural network (RNN) and long short���term memory (LSTM) network. Phys D Nonlinear Phenomena. 2020;404:132306.
  57. Abdul���Rahman HS, Gdeisat MA, Burton DR, Lalor MJ, Lilley F, Moore CJ. Fast and robust three���dimensional best path phase unwrapping algorithm. Appl. Opt. 2007;46(26):6623���6635.
  58. Sun H, Wilman AH. Background field removal using spherical mean value filtering and Tikhonov regularization. Magn Reson Med. 2014;71(3):1151���1157.
  59. Milovic C, Tejos C, Acosta���Cabronero J, et al. The 2016 QSM challenge: lessons learned and considerations for a future challenge design. Magn Reson Med. 2020;84(3):1624���1637.
  60. Milovic C, Tejos C, Silva J, Shmueli K, Irarrazaval P. XSIM: a structural similarity index measure optimized for MRI QSM. Magn Reson Med. 2025;93(1):411���421.
  61. Shi Y, Feng R, Li Z, Zhuang J, Zhang Y, Wei H. Towards in vivo ground truth susceptibility for single���orientation deep learning QSM: a multi���orientation gradient���echo MRI dataset. Neuroimage. 2022;261:119522.
  62. Marques JP, Meineke J, Milovic C, et al. QSM reconstruction challenge 2.0: a realistic in silico head phantom for MRI data simulation and evaluation of susceptibility mapping procedures. Magn Reson Med. 2021;86(1):526���542.
  63. QSM Challenge 2.0 Organization Committee. QSM reconstruction challenge 2.0: design and report of results. Magn Reson Med. 2021;86(3):1241���1255.
  64. Jung W, Bollmann S, Lee J. Overview of quantitative susceptibility mapping using deep learning: current status, challenges and opportunities. NMR Biomed. 2022;35(4):e4292.

Grants

  1. 62301616/National Natural Science Foundation of China
  2. 62301352/National Natural Science Foundation of China
  3. 2024JJ6530/Natural Science Foundation of Hunan
  4. 2021RC4008/Hunan Provincial Science and Technology Program
  5. /HighPerformance Computing Center of Central South University
  6. DE20101297/Australia Research Council
  7. DP230101628/Australia Research Council
Journal Article

Word Cloud

Created with Highcharts 10.0.0QSMresultsartifactsmethodsdeepIRQSMsusceptibilityU-netmappingnoisereconstructionlearningreverserecurrentproposediterativeLPCNNQuantitativetissuestudydipoleinversionlearning-basedaccuracyarchitecturefeaturefourconcatenationsmodulesvivoiLSQRMEDIxQSMMoDL-QSM12BACKGROUND:post-processingmagneticresonanceimagingMRItechniqueextractsdistributionsusceptibilitiesholdssignificantpromiseneurologicaldiseasesHoweverill-conditionednatureoftenfieldDeepshowngreatpotentialaddressingissueshoweverexistingapproachesrelybasicstructuresleadinglimitedperformancessometimesPURPOSE:aimsdevelopnovelmethodimprovingmitigatingleveraginguniquenetworkenhanceslatentutilizationMETHODS:advancedfeaturingiterationsmiddleoptimizefusionimprovecomparativeexperimentsbasedsimulateddatasetscarriedcomparetwotraditionalrecentlyRESULTS:workoutperformedreducingimagesachievedaveragelowestXSIM8481%simulationsshowingimprovements80%68%1866%1049%2557%1978%respectivelyyieldedleastdatapresentvisuallyappealingmeantimesuccessfullyalleviatedover-smoothingunderestimationCONCLUSION:OverallshowedsuperiorcomparedofferingaccuratesolutionclinicalapplicationsvianeuralnetworksIR2QSMquantitativemoduleconcatenation

Similar Articles

Cited By