Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study

Joshua R. Astley; Alberto M. Biancardi; Paul J. C. Hughes; Helen Marshall; Guilhem J. Collier; Ho Fung Chan; Laura C. Saunders; Laurie J. Smith; Martin L. Brook; Roger Thompson; Sarah Rowland-Jones; Sarah Skeoch; Stephen M. Bianchi; Matthew Q. Hatton; Najib M. Rahman; Ling Pei Ho; Chris E. Brightling; Louise V. Wain; Amisha Singapuri; Rachael A. Evans; Alastair J. Moss; Gerry P. McCann; Stefan Neubauer; Betty Raman; Jim M. Wild; Bilal A. Tahir

doi:10.1002/jmri.28643

Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study

, Joshua R. Astley, Alberto M. Biancardi, Paul J. C. Hughes, Helen Marshall, Guilhem J. Collier, Ho Fung Chan, Laura C. Saunders, Laurie J. Smith, Martin L. Brook, Roger Thompson, Sarah Rowland-Jones, Sarah Skeoch, Stephen M. Bianchi, Matthew Q. Hatton, Najib M. Rahman, Ling Pei Ho, Chris E. Brightling, Louise V. Wain, Amisha SingapuriRachael A. Evans, Alastair J. Moss, Gerry P. McCann, Stefan Neubauer, Betty Raman, Jim M. Wild (Lead / Corresponding author), Bilal A. Tahir

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

6 Citations (Scopus)

78 Downloads (Pure)

Abstract

Background: Recently, deep learning via convolutional neural networks (CNNs) has largely superseded conventional methods for proton (¹H)-MRI lung segmentation. However, previous deep learning studies have utilized single-center data and limited acquisition parameters.

Purpose: Develop a generalizable CNN for lung segmentation in ¹H-MRI, robust to pathology, acquisition protocol, vendor, and center.

Study type: Retrospective.

Population: A total of 809 ¹H-MRI scans from 258 participants with various pulmonary pathologies (median age (range): 57 (6–85); 42% females) and 31 healthy participants (median age (range): 34 (23–76); 34% females) that were split into training (593 scans (74%); 157 participants (55%)), testing (50 scans (6%); 50 participants (17%)) and external validation (164 scans (20%); 82 participants (28%)) sets.

Field Strength/Sequence: 1.5-T and 3-T/3D spoiled-gradient recalled and ultrashort echo-time ¹H-MRI.

Assessment: 2D and 3D CNNs, trained on single-center, multi-sequence data, and the conventional spatial fuzzy c-means (SFCM) method were compared to manually delineated expert segmentations. Each method was validated on external data originating from several centers. Dice similarity coefficient (DSC), average boundary Hausdorff distance (Average HD), and relative error (XOR) metrics to assess segmentation performance.

Statistical Tests: Kruskal–Wallis tests assessed significances of differences between acquisitions in the testing set. Friedman tests with post hoc multiple comparisons assessed differences between the 2D CNN, 3D CNN, and SFCM. Bland–Altman analyses assessed agreement with manually derived lung volumes. A P value of <0.05 was considered statistically significant.

Results: The 3D CNN significantly outperformed its 2D analog and SFCM, yielding a median (range) DSC of 0.961 (0.880–0.987), Average HD of 1.63 mm (0.65–5.45) and XOR of 0.079 (0.025–0.240) on the testing set and a DSC of 0.973 (0.866–0.987), Average HD of 1.11 mm (0.47–8.13) and XOR of 0.054 (0.026–0.255) on external validation data.

Data Conclusion: The 3D CNN generated accurate ¹H-MRI lung segmentations on a heterogenous dataset, demonstrating robustness to disease pathology, sequence, vendor, and center.

Evidence Level: 4.

Technical Efficacy: Stage 1.

Original language	English
Pages (from-to)	1030-1044
Number of pages	15
Journal	Journal of Magnetic Resonance Imaging
Volume	58
Issue number	4
Early online date	17 Feb 2023
DOIs	https://doi.org/10.1002/jmri.28643
Publication status	Published - Oct 2023

Keywords

CNN
deep learning
lung
segmentation

ASJC Scopus subject areas

Radiology Nuclear Medicine and imaging

Access to Document

10.1002/jmri.28643Licence: CC BY

Final Published VersionFinal published version, 2.01 MBLicence: CC BY

Cite this

, Astley, J. R., Biancardi, A. M., Hughes, P. J. C., Marshall, H., Collier, G. J., Chan, H. F., Saunders, L. C., Smith, L. J., Brook, M. L., Thompson, R., Rowland-Jones, S., Skeoch, S., Bianchi, S. M., Hatton, M. Q., Rahman, N. M., Ho, L. P., Brightling, C. E., Wain, L. V., ... Tahir, B. A. (2023). Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study. Journal of Magnetic Resonance Imaging, 58(4), 1030-1044. https://doi.org/10.1002/jmri.28643

@article{0adc30eb79344a3b80eed25ed9df29ed,

title = "Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study",

abstract = "Background: Recently, deep learning via convolutional neural networks (CNNs) has largely superseded conventional methods for proton (1H)-MRI lung segmentation. However, previous deep learning studies have utilized single-center data and limited acquisition parameters.Purpose: Develop a generalizable CNN for lung segmentation in 1H-MRI, robust to pathology, acquisition protocol, vendor, and center.Study type: Retrospective.Population: A total of 809 1H-MRI scans from 258 participants with various pulmonary pathologies (median age (range): 57 (6–85); 42% females) and 31 healthy participants (median age (range): 34 (23–76); 34% females) that were split into training (593 scans (74%); 157 participants (55%)), testing (50 scans (6%); 50 participants (17%)) and external validation (164 scans (20%); 82 participants (28%)) sets.Field Strength/Sequence: 1.5-T and 3-T/3D spoiled-gradient recalled and ultrashort echo-time 1H-MRI.Assessment: 2D and 3D CNNs, trained on single-center, multi-sequence data, and the conventional spatial fuzzy c-means (SFCM) method were compared to manually delineated expert segmentations. Each method was validated on external data originating from several centers. Dice similarity coefficient (DSC), average boundary Hausdorff distance (Average HD), and relative error (XOR) metrics to assess segmentation performance.Statistical Tests: Kruskal–Wallis tests assessed significances of differences between acquisitions in the testing set. Friedman tests with post hoc multiple comparisons assessed differences between the 2D CNN, 3D CNN, and SFCM. Bland–Altman analyses assessed agreement with manually derived lung volumes. A P value of <0.05 was considered statistically significant.Results: The 3D CNN significantly outperformed its 2D analog and SFCM, yielding a median (range) DSC of 0.961 (0.880–0.987), Average HD of 1.63 mm (0.65–5.45) and XOR of 0.079 (0.025–0.240) on the testing set and a DSC of 0.973 (0.866–0.987), Average HD of 1.11 mm (0.47–8.13) and XOR of 0.054 (0.026–0.255) on external validation data.Data Conclusion: The 3D CNN generated accurate 1H-MRI lung segmentations on a heterogenous dataset, demonstrating robustness to disease pathology, sequence, vendor, and center.Evidence Level: 4.Technical Efficacy: Stage 1.",

keywords = "CNN, deep learning, lung, segmentation",

author = "Astley, {Joshua R.} and Biancardi, {Alberto M.} and Hughes, {Paul J. C.} and Helen Marshall and Collier, {Guilhem J.} and Chan, {Ho Fung} and Saunders, {Laura C.} and Smith, {Laurie J.} and Brook, {Martin L.} and Roger Thompson and Sarah Rowland-Jones and Sarah Skeoch and Bianchi, {Stephen M.} and Hatton, {Matthew Q.} and Rahman, {Najib M.} and Ho, {Ling Pei} and Brightling, {Chris E.} and Wain, {Louise V.} and Amisha Singapuri and Evans, {Rachael A.} and Moss, {Alastair J.} and McCann, {Gerry P.} and Stefan Neubauer and Betty Raman and Wild, {Jim M.} and Tahir, {Bilal A.} and Jacob George",

note = "Funding Information: Grant source: This work was supported by Yorkshire Cancer Research (S406BT), National Institute of Health Research (NIHR) (NIHR‐RP‐R3‐12‐027), Medical Research Council (MR/M008894/1) and the EU Innovative Medicine Initiative (116106). C‐MORE/PHOSP‐COVID is jointly funded by a grant from the MRC‐UK Research and Innovation and the Department of Health and Social Care through the National Institute for Health Research rapid response panel to tackle COVID‐19 (MR/V027859/1 and COV0319). The views expressed in the publication are those of the author(s) and not necessarily those of the National Health Service (NHS), the NIHR or the Department of Health and Social Care. B.R is supported by BHF Oxford CRE (RE/18/3/34214). G.P.M is supported by an NIHR Professorship (NIHR‐RP‐2017‐ST2‐08‐007). A.J.M is supported by the BHF Leicester Accelerator Award. Copyright: {\textcopyright} 2023 The Authors. Journal of Magnetic Resonance Imaging published by Wiley Periodicals LLC on behalf of International Society for Magnetic Resonance in Medicine.",

year = "2023",

month = oct,

doi = "10.1002/jmri.28643",

language = "English",

volume = "58",

pages = "1030--1044",

journal = "Journal of Magnetic Resonance Imaging",

issn = "1053-1807",

publisher = "Wiley",

number = "4",

}

Astley, JR, Biancardi, AM, Hughes, PJC, Marshall, H, Collier, GJ, Chan, HF, Saunders, LC, Smith, LJ, Brook, ML, Thompson, R, Rowland-Jones, S, Skeoch, S, Bianchi, SM, Hatton, MQ, Rahman, NM, Ho, LP, Brightling, CE, Wain, LV, Singapuri, A, Evans, RA, Moss, AJ, McCann, GP, Neubauer, S, Raman, B & Wild, JM & Tahir, BA 2023, 'Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study', Journal of Magnetic Resonance Imaging, vol. 58, no. 4, pp. 1030-1044. https://doi.org/10.1002/jmri.28643

TY - JOUR

T1 - Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation

T2 - A Multi-center, Multi-vendor, and Multi-disease Study

AU - Astley, Joshua R.

AU - Biancardi, Alberto M.

AU - Hughes, Paul J. C.

AU - Marshall, Helen

AU - Collier, Guilhem J.

AU - Chan, Ho Fung

AU - Saunders, Laura C.

AU - Smith, Laurie J.

AU - Brook, Martin L.

AU - Thompson, Roger

AU - Rowland-Jones, Sarah

AU - Skeoch, Sarah

AU - Bianchi, Stephen M.

AU - Hatton, Matthew Q.

AU - Rahman, Najib M.

AU - Ho, Ling Pei

AU - Brightling, Chris E.

AU - Wain, Louise V.

AU - Singapuri, Amisha

AU - Evans, Rachael A.

AU - Moss, Alastair J.

AU - McCann, Gerry P.

AU - Neubauer, Stefan

AU - Raman, Betty

AU - Wild, Jim M.

AU - Tahir, Bilal A.

AU - George, Jacob

N1 - Funding Information: Grant source: This work was supported by Yorkshire Cancer Research (S406BT), National Institute of Health Research (NIHR) (NIHR‐RP‐R3‐12‐027), Medical Research Council (MR/M008894/1) and the EU Innovative Medicine Initiative (116106). C‐MORE/PHOSP‐COVID is jointly funded by a grant from the MRC‐UK Research and Innovation and the Department of Health and Social Care through the National Institute for Health Research rapid response panel to tackle COVID‐19 (MR/V027859/1 and COV0319). The views expressed in the publication are those of the author(s) and not necessarily those of the National Health Service (NHS), the NIHR or the Department of Health and Social Care. B.R is supported by BHF Oxford CRE (RE/18/3/34214). G.P.M is supported by an NIHR Professorship (NIHR‐RP‐2017‐ST2‐08‐007). A.J.M is supported by the BHF Leicester Accelerator Award. Copyright: © 2023 The Authors. Journal of Magnetic Resonance Imaging published by Wiley Periodicals LLC on behalf of International Society for Magnetic Resonance in Medicine.

PY - 2023/10

Y1 - 2023/10

N2 - Background: Recently, deep learning via convolutional neural networks (CNNs) has largely superseded conventional methods for proton (1H)-MRI lung segmentation. However, previous deep learning studies have utilized single-center data and limited acquisition parameters.Purpose: Develop a generalizable CNN for lung segmentation in 1H-MRI, robust to pathology, acquisition protocol, vendor, and center.Study type: Retrospective.Population: A total of 809 1H-MRI scans from 258 participants with various pulmonary pathologies (median age (range): 57 (6–85); 42% females) and 31 healthy participants (median age (range): 34 (23–76); 34% females) that were split into training (593 scans (74%); 157 participants (55%)), testing (50 scans (6%); 50 participants (17%)) and external validation (164 scans (20%); 82 participants (28%)) sets.Field Strength/Sequence: 1.5-T and 3-T/3D spoiled-gradient recalled and ultrashort echo-time 1H-MRI.Assessment: 2D and 3D CNNs, trained on single-center, multi-sequence data, and the conventional spatial fuzzy c-means (SFCM) method were compared to manually delineated expert segmentations. Each method was validated on external data originating from several centers. Dice similarity coefficient (DSC), average boundary Hausdorff distance (Average HD), and relative error (XOR) metrics to assess segmentation performance.Statistical Tests: Kruskal–Wallis tests assessed significances of differences between acquisitions in the testing set. Friedman tests with post hoc multiple comparisons assessed differences between the 2D CNN, 3D CNN, and SFCM. Bland–Altman analyses assessed agreement with manually derived lung volumes. A P value of <0.05 was considered statistically significant.Results: The 3D CNN significantly outperformed its 2D analog and SFCM, yielding a median (range) DSC of 0.961 (0.880–0.987), Average HD of 1.63 mm (0.65–5.45) and XOR of 0.079 (0.025–0.240) on the testing set and a DSC of 0.973 (0.866–0.987), Average HD of 1.11 mm (0.47–8.13) and XOR of 0.054 (0.026–0.255) on external validation data.Data Conclusion: The 3D CNN generated accurate 1H-MRI lung segmentations on a heterogenous dataset, demonstrating robustness to disease pathology, sequence, vendor, and center.Evidence Level: 4.Technical Efficacy: Stage 1.

AB - Background: Recently, deep learning via convolutional neural networks (CNNs) has largely superseded conventional methods for proton (1H)-MRI lung segmentation. However, previous deep learning studies have utilized single-center data and limited acquisition parameters.Purpose: Develop a generalizable CNN for lung segmentation in 1H-MRI, robust to pathology, acquisition protocol, vendor, and center.Study type: Retrospective.Population: A total of 809 1H-MRI scans from 258 participants with various pulmonary pathologies (median age (range): 57 (6–85); 42% females) and 31 healthy participants (median age (range): 34 (23–76); 34% females) that were split into training (593 scans (74%); 157 participants (55%)), testing (50 scans (6%); 50 participants (17%)) and external validation (164 scans (20%); 82 participants (28%)) sets.Field Strength/Sequence: 1.5-T and 3-T/3D spoiled-gradient recalled and ultrashort echo-time 1H-MRI.Assessment: 2D and 3D CNNs, trained on single-center, multi-sequence data, and the conventional spatial fuzzy c-means (SFCM) method were compared to manually delineated expert segmentations. Each method was validated on external data originating from several centers. Dice similarity coefficient (DSC), average boundary Hausdorff distance (Average HD), and relative error (XOR) metrics to assess segmentation performance.Statistical Tests: Kruskal–Wallis tests assessed significances of differences between acquisitions in the testing set. Friedman tests with post hoc multiple comparisons assessed differences between the 2D CNN, 3D CNN, and SFCM. Bland–Altman analyses assessed agreement with manually derived lung volumes. A P value of <0.05 was considered statistically significant.Results: The 3D CNN significantly outperformed its 2D analog and SFCM, yielding a median (range) DSC of 0.961 (0.880–0.987), Average HD of 1.63 mm (0.65–5.45) and XOR of 0.079 (0.025–0.240) on the testing set and a DSC of 0.973 (0.866–0.987), Average HD of 1.11 mm (0.47–8.13) and XOR of 0.054 (0.026–0.255) on external validation data.Data Conclusion: The 3D CNN generated accurate 1H-MRI lung segmentations on a heterogenous dataset, demonstrating robustness to disease pathology, sequence, vendor, and center.Evidence Level: 4.Technical Efficacy: Stage 1.

KW - CNN

KW - deep learning

KW - lung

KW - segmentation

UR - http://www.scopus.com/inward/record.url?scp=85148533564&partnerID=8YFLogxK

U2 - 10.1002/jmri.28643

DO - 10.1002/jmri.28643

M3 - Article

C2 - 36799341

AN - SCOPUS:85148533564

SN - 1053-1807

VL - 58

SP - 1030

EP - 1044

JO - Journal of Magnetic Resonance Imaging

JF - Journal of Magnetic Resonance Imaging

IS - 4

ER -

Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

PHOS-COVID (Joint with University of Leicester)

Cite this

Implementable Deep Learning for Multi-sequence Proton MRI Lung Segmentation: A Multi-center, Multi-vendor, and Multi-disease Study

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Projects

PHOS-COVID (Joint with University of Leicester)

Cite this