A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images

Qing Lyu; Sanjeev V. Namjoshi; Emory McTyre; Umit Topaloglu; Richard Barcus; Michael D. Chan; Christina K. Cramer; Waldemar Debinski; Metin N. Gurcan; Glenn J. Lesser; Hui Kuan Lin; Reginald F. Munden; Boris C. Pasche; Kiran K.S. Sai; Roy E. Strowd; Stephen B. Tatter; Kounosuke Watabe; Wei Zhang; Ge Wang; Christopher T. Whitlow

doi:10.1016/j.patter.2022.100613

A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images

Qing Lyu, Sanjeev V. Namjoshi, Emory McTyre, Umit Topaloglu, Richard Barcus, Michael D. Chan, Christina K. Cramer, Waldemar Debinski, Metin N. Gurcan, Glenn J. Lesser, Hui Kuan Lin, Reginald F. Munden, Boris C. Pasche, Kiran K.S. Sai, Roy E. Strowd, Stephen B. Tatter, Kounosuke Watabe, Wei Zhang, Ge Wang, Christopher T. Whitlow

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

17 Scopus citations

Abstract

Treatment decisions for brain metastatic disease rely on knowledge of the primary organ site and are currently made with biopsy and histology. Here, we develop a deep-learning approach for accurate non-invasive digital histology with whole-brain magnetic resonance imaging (MRI) data. Contrast-enhanced T1-weighted and fast spoiled gradient echo brain MRI exams (n = 1,582) were preprocessed and input to the proposed deep-learning workflow for tumor segmentation, modality transfer, and primary site classification into one of five classes. Tenfold cross-validation generated an overall area under the receiver operating characteristic curve (AUC) of 0.878 (95% confidence interval [CI]: 0.873,0.883). These data establish that whole-brain imaging features are discriminative enough to allow accurate diagnosis of the primary organ site of malignancy. Our end-to-end deep radiomic approach has great potential for classifying metastatic tumor types from whole-brain MRI images. Further refinement may offer an invaluable clinical tool to expedite primary cancer site identification for precision treatment and improved outcomes.

Original language	English (US)
Article number	100613
Pages (from-to)	100613
Journal	Patterns
Volume	3
Issue number	11
DOIs	https://doi.org/10.1016/j.patter.2022.100613
State	Published - Nov 11 2022

Keywords

DSML 2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem
MRI
brain metastasis
classification
deep learning
primary organ site
vision transformer

ASJC Scopus subject areas

General Decision Sciences

Access to Document

10.1016/j.patter.2022.100613

Cite this

Lyu, Q., Namjoshi, S. V., McTyre, E., Topaloglu, U., Barcus, R., Chan, M. D., Cramer, C. K., Debinski, W., Gurcan, M. N., Lesser, G. J., Lin, H. K., Munden, R. F., Pasche, B. C., Sai, K. K. S., Strowd, R. E., Tatter, S. B., Watabe, K., Zhang, W., Wang, G., & Whitlow, C. T. (2022). A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images. Patterns, 3(11), 100613. Article 100613. https://doi.org/10.1016/j.patter.2022.100613

Lyu, Q, Namjoshi, SV, McTyre, E, Topaloglu, U, Barcus, R, Chan, MD, Cramer, CK, Debinski, W, Gurcan, MN, Lesser, GJ, Lin, HK, Munden, RF, Pasche, BC, Sai, KKS, Strowd, RE, Tatter, SB, Watabe, K, Zhang, W, Wang, G & Whitlow, CT 2022, 'A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images', Patterns, vol. 3, no. 11, 100613, pp. 100613. https://doi.org/10.1016/j.patter.2022.100613

@article{4770cb73deb44eedbf48fcd2218f5eb4,

title = "A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images",

abstract = "Treatment decisions for brain metastatic disease rely on knowledge of the primary organ site and are currently made with biopsy and histology. Here, we develop a deep-learning approach for accurate non-invasive digital histology with whole-brain magnetic resonance imaging (MRI) data. Contrast-enhanced T1-weighted and fast spoiled gradient echo brain MRI exams (n = 1,582) were preprocessed and input to the proposed deep-learning workflow for tumor segmentation, modality transfer, and primary site classification into one of five classes. Tenfold cross-validation generated an overall area under the receiver operating characteristic curve (AUC) of 0.878 (95% confidence interval [CI]: 0.873,0.883). These data establish that whole-brain imaging features are discriminative enough to allow accurate diagnosis of the primary organ site of malignancy. Our end-to-end deep radiomic approach has great potential for classifying metastatic tumor types from whole-brain MRI images. Further refinement may offer an invaluable clinical tool to expedite primary cancer site identification for precision treatment and improved outcomes.",

keywords = "DSML 2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem, MRI, brain metastasis, classification, deep learning, primary organ site, vision transformer",

author = "Qing Lyu and Namjoshi, {Sanjeev V.} and Emory McTyre and Umit Topaloglu and Richard Barcus and Chan, {Michael D.} and Cramer, {Christina K.} and Waldemar Debinski and Gurcan, {Metin N.} and Lesser, {Glenn J.} and Lin, {Hui Kuan} and Munden, {Reginald F.} and Pasche, {Boris C.} and Sai, {Kiran K.S.} and Strowd, {Roy E.} and Tatter, {Stephen B.} and Kounosuke Watabe and Wei Zhang and Ge Wang and Whitlow, {Christopher T.}",

note = "Publisher Copyright: {\textcopyright} 2022 The Author(s)",

year = "2022",

month = nov,

day = "11",

doi = "10.1016/j.patter.2022.100613",

language = "English (US)",

volume = "3",

pages = "100613",

journal = "Patterns",

issn = "2666-3899",

publisher = "Cell Press",

number = "11",

}

TY - JOUR

T1 - A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images

AU - Lyu, Qing

AU - Namjoshi, Sanjeev V.

AU - McTyre, Emory

AU - Topaloglu, Umit

AU - Barcus, Richard

AU - Chan, Michael D.

AU - Cramer, Christina K.

AU - Debinski, Waldemar

AU - Gurcan, Metin N.

AU - Lesser, Glenn J.

AU - Lin, Hui Kuan

AU - Munden, Reginald F.

AU - Pasche, Boris C.

AU - Sai, Kiran K.S.

AU - Strowd, Roy E.

AU - Tatter, Stephen B.

AU - Watabe, Kounosuke

AU - Zhang, Wei

AU - Wang, Ge

AU - Whitlow, Christopher T.

PY - 2022/11/11

Y1 - 2022/11/11

N2 - Treatment decisions for brain metastatic disease rely on knowledge of the primary organ site and are currently made with biopsy and histology. Here, we develop a deep-learning approach for accurate non-invasive digital histology with whole-brain magnetic resonance imaging (MRI) data. Contrast-enhanced T1-weighted and fast spoiled gradient echo brain MRI exams (n = 1,582) were preprocessed and input to the proposed deep-learning workflow for tumor segmentation, modality transfer, and primary site classification into one of five classes. Tenfold cross-validation generated an overall area under the receiver operating characteristic curve (AUC) of 0.878 (95% confidence interval [CI]: 0.873,0.883). These data establish that whole-brain imaging features are discriminative enough to allow accurate diagnosis of the primary organ site of malignancy. Our end-to-end deep radiomic approach has great potential for classifying metastatic tumor types from whole-brain MRI images. Further refinement may offer an invaluable clinical tool to expedite primary cancer site identification for precision treatment and improved outcomes.

AB - Treatment decisions for brain metastatic disease rely on knowledge of the primary organ site and are currently made with biopsy and histology. Here, we develop a deep-learning approach for accurate non-invasive digital histology with whole-brain magnetic resonance imaging (MRI) data. Contrast-enhanced T1-weighted and fast spoiled gradient echo brain MRI exams (n = 1,582) were preprocessed and input to the proposed deep-learning workflow for tumor segmentation, modality transfer, and primary site classification into one of five classes. Tenfold cross-validation generated an overall area under the receiver operating characteristic curve (AUC) of 0.878 (95% confidence interval [CI]: 0.873,0.883). These data establish that whole-brain imaging features are discriminative enough to allow accurate diagnosis of the primary organ site of malignancy. Our end-to-end deep radiomic approach has great potential for classifying metastatic tumor types from whole-brain MRI images. Further refinement may offer an invaluable clinical tool to expedite primary cancer site identification for precision treatment and improved outcomes.

KW - DSML 2: Proof-of-concept: Data science output has been formulated, implemented, and tested for one domain/problem

KW - MRI

KW - brain metastasis

KW - classification

KW - deep learning

KW - primary organ site

KW - vision transformer

UR - http://www.scopus.com/inward/record.url?scp=85141518479&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85141518479&partnerID=8YFLogxK

U2 - 10.1016/j.patter.2022.100613

DO - 10.1016/j.patter.2022.100613

M3 - Article

C2 - 36419451

AN - SCOPUS:85141518479

SN - 2666-3899

VL - 3

SP - 100613

JO - Patterns

JF - Patterns

IS - 11

M1 - 100613

ER -

A transformer-based deep-learning approach for classifying brain metastases into primary organ sites using clinical whole-brain MRI images

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this