Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)

Alysha Chelliah; David A. Wood; Liane S. Canas; Haris Shuaib; Stuart Currie; Kavi Fatania; Russell Frood; Chris Rowland-Hill; Stefanie Thust; Stephen J. Wastling; Sean Tenant; Catherine McBain; Karen Foweraker; Matthew Williams; Qiquan Wang; Andrei Roman; Carmen Dragos; Mark MacDonald; Yue Hui Lau; Christian A. Linares; Ahmed Bassiouny; Aysha Luis; Thomas Young; Juliet Brock; Edward Chandy; Erica Beaumont; Tai Chung Lam; Liam Welsh; Joanne Lewis; Ryan Mathew; Eric Kerfoot; Richard Brown; Daniel Beasley; Jennifer Glendenning; Lucy Brazil; Angela Swampillai; Keyoumars Ashkan; Sébastien Ourselin; Marc Modat; Thomas C. Booth

doi:10.1093/neuonc/noae017

Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)

Alysha Chelliah, David A. Wood, Liane S. Canas, Haris Shuaib, Stuart Currie, Kavi Fatania, Russell Frood, Chris Rowland-Hill, Stefanie Thust, Stephen J. Wastling, Sean Tenant, Catherine McBain, Karen Foweraker, Matthew Williams, Qiquan Wang, Andrei Roman, Carmen Dragos, Mark MacDonald, Yue Hui Lau, Christian A. LinaresAhmed Bassiouny, Aysha Luis, Thomas Young, Juliet Brock, Edward Chandy, Erica Beaumont, Tai Chung Lam, Liam Welsh, Joanne Lewis, Ryan Mathew, Eric Kerfoot, Richard Brown, Daniel Beasley, Jennifer Glendenning, Lucy Brazil, Angela Swampillai, Keyoumars Ashkan, Sébastien Ourselin, Marc Modat, Thomas C. Booth^*

^*Corresponding author for this work

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

1 Citation (Scopus)

Abstract

Background. The aim was to predict survival of glioblastoma at 8 months after radiotherapy (a period allowing for completing a typical course of adjuvant temozolomide), by applying deep learning to the first brain MRI after radiotherapy completion. Methods. Retrospective and prospective data were collected from 206 consecutive glioblastoma, isocitrate dehydrogenase -wildtype patients diagnosed between March 2014 and February 2022 across 11 UK centers. Models were trained on 158 retrospective patients from 3 centers. Holdout test sets were retrospective (n = 19; internal validation), and prospective (n = 29; external validation from 8 distinct centers). Neural network branches for T₂weighted and contrast-enhanced T₁-weighted inputs were concatenated to predict survival. A nonimaging branch (demographics/MGMT/treatment data) was also combined with the imaging model. We investigated the influence of individual MR sequences; nonimaging features; and weighted dense blocks pretrained for abnormality detection. Results. The imaging model outperformed the nonimaging model in all test sets (area under the receiver-operating characteristic curve, AUC P = .038) and performed similarly to a combined imaging/nonimaging model (P > .05). Imaging, nonimaging, and combined models applied to amalgamated test sets gave AUCs of 0.93, 0.79, and 0.91. Initializing the imaging model with pretrained weights from 10 000s of brain MRIs improved performance considerably (amalgamated test sets without pretraining 0.64; P = .003). Conclusions. A deep learning model using MRI images after radiotherapy reliably and accurately determined survival of glioblastoma. The model serves as a prognostic biomarker identifying patients who will not survive beyond a typical course of adjuvant temozolomide, thereby stratifying patients into those who might require early second-line or clinical trial treatment.

Original language	English
Pages (from-to)	1138-1151
Number of pages	14
Journal	Neuro-oncology
Volume	26
Issue number	6
DOIs	https://doi.org/10.1093/neuonc/noae017
Publication status	Published - 1 Jun 2024

Keywords

artificial intelligence
deep learning
glioblastoma
magnetic resonance imaging
survival

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1093/neuonc/noae017

Cite this

Chelliah, A., Wood, D. A., Canas, L. S., Shuaib, H., Currie, S., Fatania, K., Frood, R., Rowland-Hill, C., Thust, S., Wastling, S. J., Tenant, S., McBain, C., Foweraker, K., Williams, M., Wang, Q., Roman, A., Dragos, C., MacDonald, M., Lau, Y. H., ... Booth, T. C. (2024). Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study). Neuro-oncology, 26(6), 1138-1151. https://doi.org/10.1093/neuonc/noae017

@article{8acd30561aa546a3baaece031d1f063f,

title = "Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)",

abstract = "Background. The aim was to predict survival of glioblastoma at 8 months after radiotherapy (a period allowing for completing a typical course of adjuvant temozolomide), by applying deep learning to the first brain MRI after radiotherapy completion. Methods. Retrospective and prospective data were collected from 206 consecutive glioblastoma, isocitrate dehydrogenase -wildtype patients diagnosed between March 2014 and February 2022 across 11 UK centers. Models were trained on 158 retrospective patients from 3 centers. Holdout test sets were retrospective (n = 19; internal validation), and prospective (n = 29; external validation from 8 distinct centers). Neural network branches for T2weighted and contrast-enhanced T1-weighted inputs were concatenated to predict survival. A nonimaging branch (demographics/MGMT/treatment data) was also combined with the imaging model. We investigated the influence of individual MR sequences; nonimaging features; and weighted dense blocks pretrained for abnormality detection. Results. The imaging model outperformed the nonimaging model in all test sets (area under the receiver-operating characteristic curve, AUC P = .038) and performed similarly to a combined imaging/nonimaging model (P > .05). Imaging, nonimaging, and combined models applied to amalgamated test sets gave AUCs of 0.93, 0.79, and 0.91. Initializing the imaging model with pretrained weights from 10 000s of brain MRIs improved performance considerably (amalgamated test sets without pretraining 0.64; P = .003). Conclusions. A deep learning model using MRI images after radiotherapy reliably and accurately determined survival of glioblastoma. The model serves as a prognostic biomarker identifying patients who will not survive beyond a typical course of adjuvant temozolomide, thereby stratifying patients into those who might require early second-line or clinical trial treatment.",

keywords = "artificial intelligence, deep learning, glioblastoma, magnetic resonance imaging, survival",

author = "Alysha Chelliah and Wood, {David A.} and Canas, {Liane S.} and Haris Shuaib and Stuart Currie and Kavi Fatania and Russell Frood and Chris Rowland-Hill and Stefanie Thust and Wastling, {Stephen J.} and Sean Tenant and Catherine McBain and Karen Foweraker and Matthew Williams and Qiquan Wang and Andrei Roman and Carmen Dragos and Mark MacDonald and Lau, {Yue Hui} and Linares, {Christian A.} and Ahmed Bassiouny and Aysha Luis and Thomas Young and Juliet Brock and Edward Chandy and Erica Beaumont and Lam, {Tai Chung} and Liam Welsh and Joanne Lewis and Ryan Mathew and Eric Kerfoot and Richard Brown and Daniel Beasley and Jennifer Glendenning and Lucy Brazil and Angela Swampillai and Keyoumars Ashkan and S{\'e}bastien Ourselin and Marc Modat and Booth, {Thomas C.}",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2024.",

year = "2024",

month = jun,

day = "1",

doi = "10.1093/neuonc/noae017",

language = "English",

volume = "26",

pages = "1138--1151",

journal = "Neuro-oncology",

issn = "1522-8517",

publisher = "Oxford University Press",

number = "6",

}

Chelliah, A, Wood, DA, Canas, LS, Shuaib, H, Currie, S, Fatania, K, Frood, R, Rowland-Hill, C, Thust, S, Wastling, SJ, Tenant, S, McBain, C, Foweraker, K, Williams, M, Wang, Q, Roman, A, Dragos, C, MacDonald, M, Lau, YH, Linares, CA, Bassiouny, A, Luis, A, Young, T, Brock, J, Chandy, E, Beaumont, E, Lam, TC, Welsh, L, Lewis, J, Mathew, R, Kerfoot, E, Brown, R, Beasley, D, Glendenning, J, Brazil, L, Swampillai, A, Ashkan, K , Ourselin, S , Modat, M & Booth, TC 2024, 'Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)', Neuro-oncology, vol. 26, no. 6, pp. 1138-1151. https://doi.org/10.1093/neuonc/noae017

TY - JOUR

T1 - Glioblastoma and radiotherapy

T2 - A multicenter AI study for Survival Predictions from MRI (GRASP study)

AU - Chelliah, Alysha

AU - Wood, David A.

AU - Canas, Liane S.

AU - Shuaib, Haris

AU - Currie, Stuart

AU - Fatania, Kavi

AU - Frood, Russell

AU - Rowland-Hill, Chris

AU - Thust, Stefanie

AU - Wastling, Stephen J.

AU - Tenant, Sean

AU - McBain, Catherine

AU - Foweraker, Karen

AU - Williams, Matthew

AU - Wang, Qiquan

AU - Roman, Andrei

AU - Dragos, Carmen

AU - MacDonald, Mark

AU - Lau, Yue Hui

AU - Linares, Christian A.

AU - Bassiouny, Ahmed

AU - Luis, Aysha

AU - Young, Thomas

AU - Brock, Juliet

AU - Chandy, Edward

AU - Beaumont, Erica

AU - Lam, Tai Chung

AU - Welsh, Liam

AU - Lewis, Joanne

AU - Mathew, Ryan

AU - Kerfoot, Eric

AU - Brown, Richard

AU - Beasley, Daniel

AU - Glendenning, Jennifer

AU - Brazil, Lucy

AU - Swampillai, Angela

AU - Ashkan, Keyoumars

AU - Ourselin, Sébastien

AU - Modat, Marc

AU - Booth, Thomas C.

PY - 2024/6/1

Y1 - 2024/6/1

N2 - Background. The aim was to predict survival of glioblastoma at 8 months after radiotherapy (a period allowing for completing a typical course of adjuvant temozolomide), by applying deep learning to the first brain MRI after radiotherapy completion. Methods. Retrospective and prospective data were collected from 206 consecutive glioblastoma, isocitrate dehydrogenase -wildtype patients diagnosed between March 2014 and February 2022 across 11 UK centers. Models were trained on 158 retrospective patients from 3 centers. Holdout test sets were retrospective (n = 19; internal validation), and prospective (n = 29; external validation from 8 distinct centers). Neural network branches for T2weighted and contrast-enhanced T1-weighted inputs were concatenated to predict survival. A nonimaging branch (demographics/MGMT/treatment data) was also combined with the imaging model. We investigated the influence of individual MR sequences; nonimaging features; and weighted dense blocks pretrained for abnormality detection. Results. The imaging model outperformed the nonimaging model in all test sets (area under the receiver-operating characteristic curve, AUC P = .038) and performed similarly to a combined imaging/nonimaging model (P > .05). Imaging, nonimaging, and combined models applied to amalgamated test sets gave AUCs of 0.93, 0.79, and 0.91. Initializing the imaging model with pretrained weights from 10 000s of brain MRIs improved performance considerably (amalgamated test sets without pretraining 0.64; P = .003). Conclusions. A deep learning model using MRI images after radiotherapy reliably and accurately determined survival of glioblastoma. The model serves as a prognostic biomarker identifying patients who will not survive beyond a typical course of adjuvant temozolomide, thereby stratifying patients into those who might require early second-line or clinical trial treatment.

AB - Background. The aim was to predict survival of glioblastoma at 8 months after radiotherapy (a period allowing for completing a typical course of adjuvant temozolomide), by applying deep learning to the first brain MRI after radiotherapy completion. Methods. Retrospective and prospective data were collected from 206 consecutive glioblastoma, isocitrate dehydrogenase -wildtype patients diagnosed between March 2014 and February 2022 across 11 UK centers. Models were trained on 158 retrospective patients from 3 centers. Holdout test sets were retrospective (n = 19; internal validation), and prospective (n = 29; external validation from 8 distinct centers). Neural network branches for T2weighted and contrast-enhanced T1-weighted inputs were concatenated to predict survival. A nonimaging branch (demographics/MGMT/treatment data) was also combined with the imaging model. We investigated the influence of individual MR sequences; nonimaging features; and weighted dense blocks pretrained for abnormality detection. Results. The imaging model outperformed the nonimaging model in all test sets (area under the receiver-operating characteristic curve, AUC P = .038) and performed similarly to a combined imaging/nonimaging model (P > .05). Imaging, nonimaging, and combined models applied to amalgamated test sets gave AUCs of 0.93, 0.79, and 0.91. Initializing the imaging model with pretrained weights from 10 000s of brain MRIs improved performance considerably (amalgamated test sets without pretraining 0.64; P = .003). Conclusions. A deep learning model using MRI images after radiotherapy reliably and accurately determined survival of glioblastoma. The model serves as a prognostic biomarker identifying patients who will not survive beyond a typical course of adjuvant temozolomide, thereby stratifying patients into those who might require early second-line or clinical trial treatment.

KW - artificial intelligence

KW - deep learning

KW - glioblastoma

KW - magnetic resonance imaging

KW - survival

UR - http://www.scopus.com/inward/record.url?scp=85195226563&partnerID=8YFLogxK

U2 - 10.1093/neuonc/noae017

DO - 10.1093/neuonc/noae017

M3 - Article

C2 - 38285679

AN - SCOPUS:85195226563

SN - 1522-8517

VL - 26

SP - 1138

EP - 1151

JO - Neuro-oncology

JF - Neuro-oncology

IS - 6

ER -

Glioblastoma and radiotherapy: A multicenter AI study for Survival Predictions from MRI (GRASP study)

Abstract

Keywords

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this