MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study

Lydia T. Tam; Kristen W. Yeom; Jason N. Wright; Alok Jaju; Alireza Radmanesh; Michelle Han; Sebastian Toescu; Maryam Maleki; Eric Chen; Andrew Campion; Hollie A. Lai; Azam A. Eghbal; Ozgur Oztekin; Kshitij Mankad; Darren Hargrave; Thomas S. Jacques; Robert Goetti; Robert M. Lober; Samuel H. Cheshier; Sandy Napel; Mourad Said; Kristian Aquilina; Chang Y. Ho; Michelle Monje; Nicholas A. Vitanza; Sarah A. Mattonen

doi:10.1093/noajnl/vdab042

MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study

Lydia T. Tam, Kristen W. Yeom^*, Jason N. Wright, Alok Jaju, Alireza Radmanesh, Michelle Han, Sebastian Toescu, Maryam Maleki, Eric Chen, Andrew Campion, Hollie A. Lai, Azam A. Eghbal, Ozgur Oztekin, Kshitij Mankad, Darren Hargrave, Thomas S. Jacques, Robert Goetti, Robert M. Lober, Samuel H. Cheshier, Sandy NapelMourad Said, Kristian Aquilina, Chang Y. Ho, Michelle Monje, Nicholas A. Vitanza^*, Sarah A. Mattonen^*

^*Corresponding author for this work

Radiology

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

20 Scopus citations

Abstract

Background. Diffuse intrinsic pontine gliomas (DIPGs) are lethal pediatric brain tumors. Presently, MRI is the mainstay of disease diagnosis and surveillance. We identify clinically significant computational features from MRI and create a prognostic machine learning model. Methods. We isolated tumor volumes of T1-post-contrast (T1) and T2-weighted (T2) MRIs from 177 treatment-naïve DIPG patients from an international cohort for model training and testing. The Quantitative Image Feature Pipeline and PyRadiomics was used for feature extraction. Ten-fold cross-validation of least absolute shrinkage and selection operator Cox regression selected optimal features to predict overall survival in the training dataset and tested in the independent testing dataset. We analyzed model performance using clinical variables (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variables. Results. All selected features were intensity and texture-based on the wavelet-filtered images (3 T1 graylevel co-occurrence matrix (GLCM) texture features, T2 GLCM texture feature, and T2 first-order mean). This multivariable Cox model demonstrated a concordance of 0.68 (95% CI: 0.61-0.74) in the training dataset, significantly outperforming the clinical-only model (C = 0.57 [95% CI: 0.49-0.64]). Adding clinical features to radiomics slightly improved performance (C = 0.70 [95% CI: 0.64-0.77]). The combined radiomics and clinical model was validated in the independent testing dataset (C = 0.59 [95% CI: 0.51-0.67], Noether's test P = .02). Conclusions. In this international study, we demonstrate the use of radiomic signatures to create a machine learning model for DIPG prognostication. Standardized, quantitative approaches that objectively measure DIPG changes, including computational MRI evaluation, could offer new approaches to assessing tumor phenotype and serve a future role for optimizing clinical trial eligibility and tumor surveillance.

Original language	English (US)
Article number	vdab042
Journal	Neuro-Oncology Advances
Volume	3
Issue number	1
DOIs	https://doi.org/10.1093/noajnl/vdab042
State	Published - Jan 1 2021

Keywords

H3K27M-mutant
diffuse intrinsic pontine gliomas
diffuse midline glioma
machine learning
magnetic resonance imaging
radiomics

ASJC Scopus subject areas

Clinical Neurology
Oncology
Surgery

Access to Document

10.1093/noajnl/vdab042

Cite this

Tam, L. T., Yeom, K. W., Wright, J. N., Jaju, A., Radmanesh, A., Han, M., Toescu, S., Maleki, M., Chen, E., Campion, A., Lai, H. A., Eghbal, A. A., Oztekin, O., Mankad, K., Hargrave, D., Jacques, T. S., Goetti, R., Lober, R. M., Cheshier, S. H., ... Mattonen, S. A. (2021). MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study. Neuro-Oncology Advances, 3(1), Article vdab042. https://doi.org/10.1093/noajnl/vdab042

@article{c10fcd7c342743dd93a9243240abb521,

title = "MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study",

abstract = "Background. Diffuse intrinsic pontine gliomas (DIPGs) are lethal pediatric brain tumors. Presently, MRI is the mainstay of disease diagnosis and surveillance. We identify clinically significant computational features from MRI and create a prognostic machine learning model. Methods. We isolated tumor volumes of T1-post-contrast (T1) and T2-weighted (T2) MRIs from 177 treatment-na{\"i}ve DIPG patients from an international cohort for model training and testing. The Quantitative Image Feature Pipeline and PyRadiomics was used for feature extraction. Ten-fold cross-validation of least absolute shrinkage and selection operator Cox regression selected optimal features to predict overall survival in the training dataset and tested in the independent testing dataset. We analyzed model performance using clinical variables (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variables. Results. All selected features were intensity and texture-based on the wavelet-filtered images (3 T1 graylevel co-occurrence matrix (GLCM) texture features, T2 GLCM texture feature, and T2 first-order mean). This multivariable Cox model demonstrated a concordance of 0.68 (95% CI: 0.61-0.74) in the training dataset, significantly outperforming the clinical-only model (C = 0.57 [95% CI: 0.49-0.64]). Adding clinical features to radiomics slightly improved performance (C = 0.70 [95% CI: 0.64-0.77]). The combined radiomics and clinical model was validated in the independent testing dataset (C = 0.59 [95% CI: 0.51-0.67], Noether's test P = .02). Conclusions. In this international study, we demonstrate the use of radiomic signatures to create a machine learning model for DIPG prognostication. Standardized, quantitative approaches that objectively measure DIPG changes, including computational MRI evaluation, could offer new approaches to assessing tumor phenotype and serve a future role for optimizing clinical trial eligibility and tumor surveillance.",

keywords = "H3K27M-mutant, diffuse intrinsic pontine gliomas, diffuse midline glioma, machine learning, magnetic resonance imaging, radiomics",

author = "Tam, {Lydia T.} and Yeom, {Kristen W.} and Wright, {Jason N.} and Alok Jaju and Alireza Radmanesh and Michelle Han and Sebastian Toescu and Maryam Maleki and Eric Chen and Andrew Campion and Lai, {Hollie A.} and Eghbal, {Azam A.} and Ozgur Oztekin and Kshitij Mankad and Darren Hargrave and Jacques, {Thomas S.} and Robert Goetti and Lober, {Robert M.} and Cheshier, {Samuel H.} and Sandy Napel and Mourad Said and Kristian Aquilina and Ho, {Chang Y.} and Michelle Monje and Vitanza, {Nicholas A.} and Mattonen, {Sarah A.}",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2021.",

year = "2021",

month = jan,

day = "1",

doi = "10.1093/noajnl/vdab042",

language = "English (US)",

volume = "3",

journal = "Neuro-Oncology Advances",

issn = "2632-2498",

publisher = "Oxford University Press",

number = "1",

}

Tam, LT, Yeom, KW, Wright, JN, Jaju, A, Radmanesh, A, Han, M, Toescu, S, Maleki, M, Chen, E, Campion, A, Lai, HA, Eghbal, AA, Oztekin, O, Mankad, K, Hargrave, D, Jacques, TS, Goetti, R, Lober, RM, Cheshier, SH, Napel, S, Said, M, Aquilina, K, Ho, CY, Monje, M, Vitanza, NA & Mattonen, SA 2021, 'MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study', Neuro-Oncology Advances, vol. 3, no. 1, vdab042. https://doi.org/10.1093/noajnl/vdab042

TY - JOUR

T1 - MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma

T2 - An international study

AU - Tam, Lydia T.

AU - Yeom, Kristen W.

AU - Wright, Jason N.

AU - Jaju, Alok

AU - Radmanesh, Alireza

AU - Han, Michelle

AU - Toescu, Sebastian

AU - Maleki, Maryam

AU - Chen, Eric

AU - Campion, Andrew

AU - Lai, Hollie A.

AU - Eghbal, Azam A.

AU - Oztekin, Ozgur

AU - Mankad, Kshitij

AU - Hargrave, Darren

AU - Jacques, Thomas S.

AU - Goetti, Robert

AU - Lober, Robert M.

AU - Cheshier, Samuel H.

AU - Napel, Sandy

AU - Said, Mourad

AU - Aquilina, Kristian

AU - Ho, Chang Y.

AU - Monje, Michelle

AU - Vitanza, Nicholas A.

AU - Mattonen, Sarah A.

PY - 2021/1/1

Y1 - 2021/1/1

N2 - Background. Diffuse intrinsic pontine gliomas (DIPGs) are lethal pediatric brain tumors. Presently, MRI is the mainstay of disease diagnosis and surveillance. We identify clinically significant computational features from MRI and create a prognostic machine learning model. Methods. We isolated tumor volumes of T1-post-contrast (T1) and T2-weighted (T2) MRIs from 177 treatment-naïve DIPG patients from an international cohort for model training and testing. The Quantitative Image Feature Pipeline and PyRadiomics was used for feature extraction. Ten-fold cross-validation of least absolute shrinkage and selection operator Cox regression selected optimal features to predict overall survival in the training dataset and tested in the independent testing dataset. We analyzed model performance using clinical variables (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variables. Results. All selected features were intensity and texture-based on the wavelet-filtered images (3 T1 graylevel co-occurrence matrix (GLCM) texture features, T2 GLCM texture feature, and T2 first-order mean). This multivariable Cox model demonstrated a concordance of 0.68 (95% CI: 0.61-0.74) in the training dataset, significantly outperforming the clinical-only model (C = 0.57 [95% CI: 0.49-0.64]). Adding clinical features to radiomics slightly improved performance (C = 0.70 [95% CI: 0.64-0.77]). The combined radiomics and clinical model was validated in the independent testing dataset (C = 0.59 [95% CI: 0.51-0.67], Noether's test P = .02). Conclusions. In this international study, we demonstrate the use of radiomic signatures to create a machine learning model for DIPG prognostication. Standardized, quantitative approaches that objectively measure DIPG changes, including computational MRI evaluation, could offer new approaches to assessing tumor phenotype and serve a future role for optimizing clinical trial eligibility and tumor surveillance.

AB - Background. Diffuse intrinsic pontine gliomas (DIPGs) are lethal pediatric brain tumors. Presently, MRI is the mainstay of disease diagnosis and surveillance. We identify clinically significant computational features from MRI and create a prognostic machine learning model. Methods. We isolated tumor volumes of T1-post-contrast (T1) and T2-weighted (T2) MRIs from 177 treatment-naïve DIPG patients from an international cohort for model training and testing. The Quantitative Image Feature Pipeline and PyRadiomics was used for feature extraction. Ten-fold cross-validation of least absolute shrinkage and selection operator Cox regression selected optimal features to predict overall survival in the training dataset and tested in the independent testing dataset. We analyzed model performance using clinical variables (age at diagnosis and sex) only, radiomics only, and radiomics plus clinical variables. Results. All selected features were intensity and texture-based on the wavelet-filtered images (3 T1 graylevel co-occurrence matrix (GLCM) texture features, T2 GLCM texture feature, and T2 first-order mean). This multivariable Cox model demonstrated a concordance of 0.68 (95% CI: 0.61-0.74) in the training dataset, significantly outperforming the clinical-only model (C = 0.57 [95% CI: 0.49-0.64]). Adding clinical features to radiomics slightly improved performance (C = 0.70 [95% CI: 0.64-0.77]). The combined radiomics and clinical model was validated in the independent testing dataset (C = 0.59 [95% CI: 0.51-0.67], Noether's test P = .02). Conclusions. In this international study, we demonstrate the use of radiomic signatures to create a machine learning model for DIPG prognostication. Standardized, quantitative approaches that objectively measure DIPG changes, including computational MRI evaluation, could offer new approaches to assessing tumor phenotype and serve a future role for optimizing clinical trial eligibility and tumor surveillance.

KW - H3K27M-mutant

KW - diffuse intrinsic pontine gliomas

KW - diffuse midline glioma

KW - machine learning

KW - magnetic resonance imaging

KW - radiomics

UR - http://www.scopus.com/inward/record.url?scp=85119104101&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85119104101&partnerID=8YFLogxK

U2 - 10.1093/noajnl/vdab042

DO - 10.1093/noajnl/vdab042

M3 - Article

C2 - 33977272

AN - SCOPUS:85119104101

SN - 2632-2498

VL - 3

JO - Neuro-Oncology Advances

JF - Neuro-Oncology Advances

IS - 1

M1 - vdab042

ER -

MRI-based radiomics for prognosis of pediatric diffuse intrinsic pontine glioma: An international study

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this