Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning

Leland S Hu, H. Yoon, J. M. Eschbacher, L. C. Baxter, Amylou Dueck, A. Nespodzany, K. A. Smith, P. Nakaji, Y. Xu, L. Wang, J. P. Karis, A. J. Hawkins-Daarud, K. W. Singleton, P. R. Jackson, B. J. Anderies, Bernard Bendok, R. S. Zimmerman, C. Quarles, A. B. Porter-Umphrey, Maciej MrugalaA. Sharma, J. M. Hoxworth, M. G. Sattur, N. Sanai, P. E. Koulemberis, C. Krishna, J. R. Mitchell, T. Wu, Nhan Tran, Kristin Swanson, J. Li

Research output: Contribution to journalArticle

Abstract

BACKGROUND AND PURPOSE: MR imaging–based modeling of tumor cell density can substantially improve targeted treatment of glioblastoma. Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a transfer learning method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient’s own histologic data. MATERIALS AND METHODS: We recruited patients with primary glioblastoma undergoing image-guided biopsies and preoperative imaging, including contrast-enhanced MR imaging, dynamic susceptibility contrast MR imaging, and diffusion tensor imaging. We calculated relative cerebral blood volume from DSC-MR imaging and mean diffusivity and fractional anisotropy from DTI. Following image coregistration, we assessed tumor cell density for each biopsy and identified corresponding localized MR imaging measurements. We then explored a range of univariate and multivariate predictive models of tumor cell density based on MR imaging measurements in a generalized one-model-fits-all approach. We then implemented both univariate and multivariate individualized transfer learning predictive models, which harness the available population-level data but allow individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized transfer learning and generalized one-model-fits-all models. RESULTS: Tumor cell density significantly correlated with relative CBV (r 0.33, P .001), and T1-weighted postcontrast (r 0.36, P .001) on univariate analysis after correcting for multiple comparisons. With single-variable modeling (using relative CBV), transfer learning increased predictive performance (r 0.53, mean absolute error 15.19%) compared with one-model-fits-all (r 0.27, mean absolute error 17.79%). With multivariate modeling, transfer learning further improved performance (r 0.88, mean absolute error 5.66%) compared with one-model-fits-all (r 0.39, mean absolute error 16.55%). CONCLUSIONS: Transfer learning significantly improves predictive modeling performance for quantifying tumor cell density in glioblastoma.

Original languageEnglish (US)
Pages (from-to)418-425
Number of pages8
JournalAmerican Journal of Neuroradiology
Volume40
Issue number3
DOIs
StatePublished - Jan 1 2019

Fingerprint

Glioblastoma
Cell Count
Neoplasms
Image-Guided Biopsy
Diffusion Tensor Imaging
Aptitude
Anisotropy
Population
Machine Learning
Transfer (Psychology)
Biopsy

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging
  • Clinical Neurology

Cite this

Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning. / Hu, Leland S; Yoon, H.; Eschbacher, J. M.; Baxter, L. C.; Dueck, Amylou; Nespodzany, A.; Smith, K. A.; Nakaji, P.; Xu, Y.; Wang, L.; Karis, J. P.; Hawkins-Daarud, A. J.; Singleton, K. W.; Jackson, P. R.; Anderies, B. J.; Bendok, Bernard; Zimmerman, R. S.; Quarles, C.; Porter-Umphrey, A. B.; Mrugala, Maciej; Sharma, A.; Hoxworth, J. M.; Sattur, M. G.; Sanai, N.; Koulemberis, P. E.; Krishna, C.; Mitchell, J. R.; Wu, T.; Tran, Nhan; Swanson, Kristin; Li, J.

In: American Journal of Neuroradiology, Vol. 40, No. 3, 01.01.2019, p. 418-425.

Research output: Contribution to journalArticle

Hu, LS, Yoon, H, Eschbacher, JM, Baxter, LC, Dueck, A, Nespodzany, A, Smith, KA, Nakaji, P, Xu, Y, Wang, L, Karis, JP, Hawkins-Daarud, AJ, Singleton, KW, Jackson, PR, Anderies, BJ, Bendok, B, Zimmerman, RS, Quarles, C, Porter-Umphrey, AB, Mrugala, M, Sharma, A, Hoxworth, JM, Sattur, MG, Sanai, N, Koulemberis, PE, Krishna, C, Mitchell, JR, Wu, T, Tran, N, Swanson, K & Li, J 2019, 'Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning', American Journal of Neuroradiology, vol. 40, no. 3, pp. 418-425. https://doi.org/10.3174/ajnr.A5981
Hu, Leland S ; Yoon, H. ; Eschbacher, J. M. ; Baxter, L. C. ; Dueck, Amylou ; Nespodzany, A. ; Smith, K. A. ; Nakaji, P. ; Xu, Y. ; Wang, L. ; Karis, J. P. ; Hawkins-Daarud, A. J. ; Singleton, K. W. ; Jackson, P. R. ; Anderies, B. J. ; Bendok, Bernard ; Zimmerman, R. S. ; Quarles, C. ; Porter-Umphrey, A. B. ; Mrugala, Maciej ; Sharma, A. ; Hoxworth, J. M. ; Sattur, M. G. ; Sanai, N. ; Koulemberis, P. E. ; Krishna, C. ; Mitchell, J. R. ; Wu, T. ; Tran, Nhan ; Swanson, Kristin ; Li, J. / Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning. In: American Journal of Neuroradiology. 2019 ; Vol. 40, No. 3. pp. 418-425.
@article{112edd98d76845ddbae9d2b5c26ef985,
title = "Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning",
abstract = "BACKGROUND AND PURPOSE: MR imaging–based modeling of tumor cell density can substantially improve targeted treatment of glioblastoma. Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a transfer learning method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient’s own histologic data. MATERIALS AND METHODS: We recruited patients with primary glioblastoma undergoing image-guided biopsies and preoperative imaging, including contrast-enhanced MR imaging, dynamic susceptibility contrast MR imaging, and diffusion tensor imaging. We calculated relative cerebral blood volume from DSC-MR imaging and mean diffusivity and fractional anisotropy from DTI. Following image coregistration, we assessed tumor cell density for each biopsy and identified corresponding localized MR imaging measurements. We then explored a range of univariate and multivariate predictive models of tumor cell density based on MR imaging measurements in a generalized one-model-fits-all approach. We then implemented both univariate and multivariate individualized transfer learning predictive models, which harness the available population-level data but allow individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized transfer learning and generalized one-model-fits-all models. RESULTS: Tumor cell density significantly correlated with relative CBV (r 0.33, P .001), and T1-weighted postcontrast (r 0.36, P .001) on univariate analysis after correcting for multiple comparisons. With single-variable modeling (using relative CBV), transfer learning increased predictive performance (r 0.53, mean absolute error 15.19{\%}) compared with one-model-fits-all (r 0.27, mean absolute error 17.79{\%}). With multivariate modeling, transfer learning further improved performance (r 0.88, mean absolute error 5.66{\%}) compared with one-model-fits-all (r 0.39, mean absolute error 16.55{\%}). CONCLUSIONS: Transfer learning significantly improves predictive modeling performance for quantifying tumor cell density in glioblastoma.",
author = "Hu, {Leland S} and H. Yoon and Eschbacher, {J. M.} and Baxter, {L. C.} and Amylou Dueck and A. Nespodzany and Smith, {K. A.} and P. Nakaji and Y. Xu and L. Wang and Karis, {J. P.} and Hawkins-Daarud, {A. J.} and Singleton, {K. W.} and Jackson, {P. R.} and Anderies, {B. J.} and Bernard Bendok and Zimmerman, {R. S.} and C. Quarles and Porter-Umphrey, {A. B.} and Maciej Mrugala and A. Sharma and Hoxworth, {J. M.} and Sattur, {M. G.} and N. Sanai and Koulemberis, {P. E.} and C. Krishna and Mitchell, {J. R.} and T. Wu and Nhan Tran and Kristin Swanson and J. Li",
year = "2019",
month = "1",
day = "1",
doi = "10.3174/ajnr.A5981",
language = "English (US)",
volume = "40",
pages = "418--425",
journal = "American Journal of Neuroradiology",
issn = "0195-6108",
publisher = "American Society of Neuroradiology",
number = "3",

}

TY - JOUR

T1 - Accurate patient-specific machine learning models of glioblastoma invasion using transfer learning

AU - Hu, Leland S

AU - Yoon, H.

AU - Eschbacher, J. M.

AU - Baxter, L. C.

AU - Dueck, Amylou

AU - Nespodzany, A.

AU - Smith, K. A.

AU - Nakaji, P.

AU - Xu, Y.

AU - Wang, L.

AU - Karis, J. P.

AU - Hawkins-Daarud, A. J.

AU - Singleton, K. W.

AU - Jackson, P. R.

AU - Anderies, B. J.

AU - Bendok, Bernard

AU - Zimmerman, R. S.

AU - Quarles, C.

AU - Porter-Umphrey, A. B.

AU - Mrugala, Maciej

AU - Sharma, A.

AU - Hoxworth, J. M.

AU - Sattur, M. G.

AU - Sanai, N.

AU - Koulemberis, P. E.

AU - Krishna, C.

AU - Mitchell, J. R.

AU - Wu, T.

AU - Tran, Nhan

AU - Swanson, Kristin

AU - Li, J.

PY - 2019/1/1

Y1 - 2019/1/1

N2 - BACKGROUND AND PURPOSE: MR imaging–based modeling of tumor cell density can substantially improve targeted treatment of glioblastoma. Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a transfer learning method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient’s own histologic data. MATERIALS AND METHODS: We recruited patients with primary glioblastoma undergoing image-guided biopsies and preoperative imaging, including contrast-enhanced MR imaging, dynamic susceptibility contrast MR imaging, and diffusion tensor imaging. We calculated relative cerebral blood volume from DSC-MR imaging and mean diffusivity and fractional anisotropy from DTI. Following image coregistration, we assessed tumor cell density for each biopsy and identified corresponding localized MR imaging measurements. We then explored a range of univariate and multivariate predictive models of tumor cell density based on MR imaging measurements in a generalized one-model-fits-all approach. We then implemented both univariate and multivariate individualized transfer learning predictive models, which harness the available population-level data but allow individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized transfer learning and generalized one-model-fits-all models. RESULTS: Tumor cell density significantly correlated with relative CBV (r 0.33, P .001), and T1-weighted postcontrast (r 0.36, P .001) on univariate analysis after correcting for multiple comparisons. With single-variable modeling (using relative CBV), transfer learning increased predictive performance (r 0.53, mean absolute error 15.19%) compared with one-model-fits-all (r 0.27, mean absolute error 17.79%). With multivariate modeling, transfer learning further improved performance (r 0.88, mean absolute error 5.66%) compared with one-model-fits-all (r 0.39, mean absolute error 16.55%). CONCLUSIONS: Transfer learning significantly improves predictive modeling performance for quantifying tumor cell density in glioblastoma.

AB - BACKGROUND AND PURPOSE: MR imaging–based modeling of tumor cell density can substantially improve targeted treatment of glioblastoma. Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a transfer learning method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient’s own histologic data. MATERIALS AND METHODS: We recruited patients with primary glioblastoma undergoing image-guided biopsies and preoperative imaging, including contrast-enhanced MR imaging, dynamic susceptibility contrast MR imaging, and diffusion tensor imaging. We calculated relative cerebral blood volume from DSC-MR imaging and mean diffusivity and fractional anisotropy from DTI. Following image coregistration, we assessed tumor cell density for each biopsy and identified corresponding localized MR imaging measurements. We then explored a range of univariate and multivariate predictive models of tumor cell density based on MR imaging measurements in a generalized one-model-fits-all approach. We then implemented both univariate and multivariate individualized transfer learning predictive models, which harness the available population-level data but allow individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized transfer learning and generalized one-model-fits-all models. RESULTS: Tumor cell density significantly correlated with relative CBV (r 0.33, P .001), and T1-weighted postcontrast (r 0.36, P .001) on univariate analysis after correcting for multiple comparisons. With single-variable modeling (using relative CBV), transfer learning increased predictive performance (r 0.53, mean absolute error 15.19%) compared with one-model-fits-all (r 0.27, mean absolute error 17.79%). With multivariate modeling, transfer learning further improved performance (r 0.88, mean absolute error 5.66%) compared with one-model-fits-all (r 0.39, mean absolute error 16.55%). CONCLUSIONS: Transfer learning significantly improves predictive modeling performance for quantifying tumor cell density in glioblastoma.

UR - http://www.scopus.com/inward/record.url?scp=85063005062&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85063005062&partnerID=8YFLogxK

U2 - 10.3174/ajnr.A5981

DO - 10.3174/ajnr.A5981

M3 - Article

C2 - 30819771

AN - SCOPUS:85063005062

VL - 40

SP - 418

EP - 425

JO - American Journal of Neuroradiology

JF - American Journal of Neuroradiology

SN - 0195-6108

IS - 3

ER -