Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net

Khula SA Study Team; Levente Baljer; Yiqi Zhang; Niall J. Bourke; Kirsten A. Donald; Layla E. Bradford; Jessica E. Ringshaw; Simone R. Williams; Sean C.L. Deoni; Steven C.R. Williams; František Váša; Rosalyn J. Moran; Michal R. Zieff; Donna Herr; Chloë A. Jacobs; Sadeeka Williams; Zamazimba Madi; Nwabisa Mlandu; Tembeka Mhlakwaphalwa; Lauren Davel; Reese Samuels; Zayaan Goolam; Thandeka Mazubane; Bokang Methola; Khanyisa Nkubungu; Candice Knipe

doi:10.1002/hbm.70112

Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net

Khula SA Study Team, Levente Baljer^*, Yiqi Zhang, Niall J. Bourke, Kirsten A. Donald, Layla E. Bradford, Jessica E. Ringshaw, Simone R. Williams, Sean C.L. Deoni, Steven C.R. Williams, František Váša, Rosalyn J. Moran, Michal R. Zieff, Donna Herr, Chloë A. Jacobs, Sadeeka Williams, Zamazimba Madi, Nwabisa Mlandu, Tembeka Mhlakwaphalwa, Lauren DavelReese Samuels, Zayaan Goolam, Thandeka Mazubane, Bokang Methola, Khanyisa Nkubungu, Candice Knipe

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Owing to the high cost of modern magnetic resonance imaging (MRI) systems, their use in clinical care and neurodevelopmental research is limited to hospitals and universities in high income countries. Ultra-low-field systems with significantly lower scanning costs present a promising avenue towards global MRI accessibility; however, their reduced SNR compared to 1.5 or 3 T systems limits their applicability for research and clinical use. In this paper, we describe a deep learning-based super-resolution approach to generate high-resolution isotropic T₂-weighted scans from low-resolution paediatric input scans. We train a ‘multi-orientation U-Net’, which uses multiple low-resolution anisotropic images acquired in orthogonal orientations to construct a super-resolved output. Our approach exhibits improved quality of outputs compared to current state-of-the-art methods for super-resolution of ultra-low-field scans in paediatric populations. Crucially for paediatric development, our approach improves reconstruction of deep brain structures with the greatest improvement in volume estimates of the caudate, where our model improves upon the state-of-the-art in: linear correlation (r = 0.94 vs. 0.84 using existing methods), exact agreement (Lin's concordance correlation = 0.94 vs. 0.80) and mean error (0.05 cm³ vs. 0.36 cm³). Our research serves as proof-of-principle of the viability of training deep-learning based super-resolution models for use in neurodevelopmental research and presents the first model trained exclusively on paired ultra-low-field and high-field data from infants.

Original language	English
Article number	e70112
Journal	Human Brain Mapping
Volume	46
Issue number	1
Early online date	30 Dec 2024
DOIs	https://doi.org/10.1002/hbm.70112
Publication status	Published - Jan 2025

Access to Document

10.1002/hbm.70112Licence: CC BY

Cite this

Khula SA Study Team, Baljer, L., Zhang, Y., Bourke, N. J., Donald, K. A., Bradford, L. E., Ringshaw, J. E., Williams, S. R., Deoni, S. C. L., Williams, S. C. R., Váša, F., Moran, R. J., Zieff, M. R., Herr, D., Jacobs, C. A., Williams, S., Madi, Z., Mlandu, N., Mhlakwaphalwa, T., ... Knipe, C. (2025). Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net. Human Brain Mapping, 46(1), Article e70112. https://doi.org/10.1002/hbm.70112

@article{820c4479c7544d009ca241285cac8be6,

title = "Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net",

abstract = "Owing to the high cost of modern magnetic resonance imaging (MRI) systems, their use in clinical care and neurodevelopmental research is limited to hospitals and universities in high income countries. Ultra-low-field systems with significantly lower scanning costs present a promising avenue towards global MRI accessibility; however, their reduced SNR compared to 1.5 or 3 T systems limits their applicability for research and clinical use. In this paper, we describe a deep learning-based super-resolution approach to generate high-resolution isotropic T2-weighted scans from low-resolution paediatric input scans. We train a {\textquoteleft}multi-orientation U-Net{\textquoteright}, which uses multiple low-resolution anisotropic images acquired in orthogonal orientations to construct a super-resolved output. Our approach exhibits improved quality of outputs compared to current state-of-the-art methods for super-resolution of ultra-low-field scans in paediatric populations. Crucially for paediatric development, our approach improves reconstruction of deep brain structures with the greatest improvement in volume estimates of the caudate, where our model improves upon the state-of-the-art in: linear correlation (r = 0.94 vs. 0.84 using existing methods), exact agreement (Lin's concordance correlation = 0.94 vs. 0.80) and mean error (0.05 cm3 vs. 0.36 cm3). Our research serves as proof-of-principle of the viability of training deep-learning based super-resolution models for use in neurodevelopmental research and presents the first model trained exclusively on paired ultra-low-field and high-field data from infants.",

author = "{Khula SA Study Team} and Levente Baljer and Yiqi Zhang and Bourke, {Niall J.} and Donald, {Kirsten A.} and Bradford, {Layla E.} and Ringshaw, {Jessica E.} and Williams, {Simone R.} and Deoni, {Sean C.L.} and Williams, {Steven C.R.} and Franti{\v s}ek V{\'a}{\v s}a and Moran, {Rosalyn J.} and Zieff, {Michal R.} and Donna Herr and Jacobs, {Chlo{\"e} A.} and Sadeeka Williams and Zamazimba Madi and Nwabisa Mlandu and Tembeka Mhlakwaphalwa and Lauren Davel and Reese Samuels and Zayaan Goolam and Thandeka Mazubane and Bokang Methola and Khanyisa Nkubungu and Candice Knipe",

note = "Publisher Copyright: {\textcopyright} 2024 The Author(s). Human Brain Mapping published by Wiley Periodicals LLC.",

year = "2025",

month = jan,

doi = "10.1002/hbm.70112",

language = "English",

volume = "46",

journal = "Human Brain Mapping",

issn = "1065-9471",

publisher = "Wiley-Liss Inc.",

number = "1",

}

Khula SA Study Team, Baljer, L, Zhang, Y, Bourke, NJ, Donald, KA, Bradford, LE, Ringshaw, JE, Williams, SR, Deoni, SCL , Williams, SCR , Váša, F , Moran, RJ, Zieff, MR, Herr, D, Jacobs, CA, Williams, S, Madi, Z, Mlandu, N, Mhlakwaphalwa, T, Davel, L, Samuels, R, Goolam, Z, Mazubane, T, Methola, B, Nkubungu, K & Knipe, C 2025, 'Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net', Human Brain Mapping, vol. 46, no. 1, e70112. https://doi.org/10.1002/hbm.70112

TY - JOUR

T1 - Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries

T2 - Super-Resolution Using a Multi-Orientation U-Net

AU - Khula SA Study Team

AU - Baljer, Levente

AU - Zhang, Yiqi

AU - Bourke, Niall J.

AU - Donald, Kirsten A.

AU - Bradford, Layla E.

AU - Ringshaw, Jessica E.

AU - Williams, Simone R.

AU - Deoni, Sean C.L.

AU - Williams, Steven C.R.

AU - Váša, František

AU - Moran, Rosalyn J.

AU - Zieff, Michal R.

AU - Herr, Donna

AU - Jacobs, Chloë A.

AU - Williams, Sadeeka

AU - Madi, Zamazimba

AU - Mlandu, Nwabisa

AU - Mhlakwaphalwa, Tembeka

AU - Davel, Lauren

AU - Samuels, Reese

AU - Goolam, Zayaan

AU - Mazubane, Thandeka

AU - Methola, Bokang

AU - Nkubungu, Khanyisa

AU - Knipe, Candice

PY - 2025/1

Y1 - 2025/1

N2 - Owing to the high cost of modern magnetic resonance imaging (MRI) systems, their use in clinical care and neurodevelopmental research is limited to hospitals and universities in high income countries. Ultra-low-field systems with significantly lower scanning costs present a promising avenue towards global MRI accessibility; however, their reduced SNR compared to 1.5 or 3 T systems limits their applicability for research and clinical use. In this paper, we describe a deep learning-based super-resolution approach to generate high-resolution isotropic T2-weighted scans from low-resolution paediatric input scans. We train a ‘multi-orientation U-Net’, which uses multiple low-resolution anisotropic images acquired in orthogonal orientations to construct a super-resolved output. Our approach exhibits improved quality of outputs compared to current state-of-the-art methods for super-resolution of ultra-low-field scans in paediatric populations. Crucially for paediatric development, our approach improves reconstruction of deep brain structures with the greatest improvement in volume estimates of the caudate, where our model improves upon the state-of-the-art in: linear correlation (r = 0.94 vs. 0.84 using existing methods), exact agreement (Lin's concordance correlation = 0.94 vs. 0.80) and mean error (0.05 cm3 vs. 0.36 cm3). Our research serves as proof-of-principle of the viability of training deep-learning based super-resolution models for use in neurodevelopmental research and presents the first model trained exclusively on paired ultra-low-field and high-field data from infants.

AB - Owing to the high cost of modern magnetic resonance imaging (MRI) systems, their use in clinical care and neurodevelopmental research is limited to hospitals and universities in high income countries. Ultra-low-field systems with significantly lower scanning costs present a promising avenue towards global MRI accessibility; however, their reduced SNR compared to 1.5 or 3 T systems limits their applicability for research and clinical use. In this paper, we describe a deep learning-based super-resolution approach to generate high-resolution isotropic T2-weighted scans from low-resolution paediatric input scans. We train a ‘multi-orientation U-Net’, which uses multiple low-resolution anisotropic images acquired in orthogonal orientations to construct a super-resolved output. Our approach exhibits improved quality of outputs compared to current state-of-the-art methods for super-resolution of ultra-low-field scans in paediatric populations. Crucially for paediatric development, our approach improves reconstruction of deep brain structures with the greatest improvement in volume estimates of the caudate, where our model improves upon the state-of-the-art in: linear correlation (r = 0.94 vs. 0.84 using existing methods), exact agreement (Lin's concordance correlation = 0.94 vs. 0.80) and mean error (0.05 cm3 vs. 0.36 cm3). Our research serves as proof-of-principle of the viability of training deep-learning based super-resolution models for use in neurodevelopmental research and presents the first model trained exclusively on paired ultra-low-field and high-field data from infants.

UR - http://www.scopus.com/inward/record.url?scp=85214386361&partnerID=8YFLogxK

U2 - 10.1002/hbm.70112

DO - 10.1002/hbm.70112

M3 - Article

AN - SCOPUS:85214386361

SN - 1065-9471

VL - 46

JO - Human Brain Mapping

JF - Human Brain Mapping

IS - 1

M1 - e70112

ER -

Ultra-Low-Field Paediatric MRI in Low- and Middle-Income Countries: Super-Resolution Using a Multi-Orientation U-Net

Abstract

Access to Document

Other files and links

Fingerprint

Cite this