Predictive modeling of brain tumor laser ablation dynamics

Walter J. Jermakowicz, Anil K. Mahavadi, Iahn Cajigas, Lia Dan, Santiago Guerra, Ghulam Farooq, Ashish H. Shah, Pierre F. D’Haese, Michael E. Ivan, Jonathan R. Jagid, Ricardo J. Komotar

Research output: Contribution to journalArticle

Abstract

Introduction: Laser interstitial thermal therapy (LITT) is a novel MR thermometry-guided thermoablative tool revolutionizing the clinical management of brain tumors. A limitation of LITT is our inability to estimate a priori how tissues will respond to thermal energy, which hinders treatment planning and delivery. The aim of this study was to determine whether brain tumor LITT ablation dynamics may be predicted by features of the preoperative MRI and the relevance of these data, if any, to the recurrence of metastases after LITT. Methods: Intraoperative thermal damage estimate (TDE) map pixels representative of irreversible damage were retrospectively quantified relative to ablation onset for 101 LITT procedures. Raw TDE pixel counts and TDE pixel counts modelled with first order dynamics were related to eleven independent variables derived from the preoperative MRI, demographics, laser settings, and tumor pathology. Stepwise regression analysis generated predictive models of LITT dynamics, and leave-one-out cross validation evaluated the accuracy of these models at predicting TDE pixel counts solely from the independent variables. Using a deformable atlas, TDE maps were co-registered to the immediate post-ablation MRI, allowing comparison of predicted and actual ablation sizes. Results: Brain tumor TDE pixel counts modelled with first order dynamics, but not raw pixel counts, are correlated with the independent variables. Independent variables showing strong relations to the TDE pixel measures include T1 gadolinium and T2 signal, perfusion, and laser power. Associations with tissue histopathology are minimal. Leave-one-out analysis demonstrates that predictive models using these independent variables account for 77% of the variance observed in TDE pixel counts. Analysis of metastases treated revealed a trend towards the over-estimation of LITT effects by TDE maps during rapid ablations, which was associated with tumor recurrence. Conclusions: Features of the preoperative MRI are predictive of LITT ablation dynamics and could eventually be used to improve the clinical efficacy with which LITT is delivered to brain tumors.

Original languageEnglish (US)
JournalJournal of neuro-oncology
DOIs
StatePublished - Jan 1 2019

Fingerprint

Laser Therapy
Brain Neoplasms
Hot Temperature
Lasers
Therapeutics
Thermometry
Neoplasm Metastasis
Recurrence
Atlases
Gadolinium

Keywords

  • Coagulation
  • Dynamics
  • Perfusion imaging
  • Predictive modelling
  • Thermoablation

ASJC Scopus subject areas

  • Oncology
  • Neurology
  • Clinical Neurology
  • Cancer Research

Cite this

Predictive modeling of brain tumor laser ablation dynamics. / Jermakowicz, Walter J.; Mahavadi, Anil K.; Cajigas, Iahn; Dan, Lia; Guerra, Santiago; Farooq, Ghulam; Shah, Ashish H.; D’Haese, Pierre F.; Ivan, Michael E.; Jagid, Jonathan R.; Komotar, Ricardo J.

In: Journal of neuro-oncology, 01.01.2019.

Research output: Contribution to journalArticle

Jermakowicz, WJ, Mahavadi, AK, Cajigas, I, Dan, L, Guerra, S, Farooq, G, Shah, AH, D’Haese, PF, Ivan, ME, Jagid, JR & Komotar, RJ 2019, 'Predictive modeling of brain tumor laser ablation dynamics', Journal of neuro-oncology. https://doi.org/10.1007/s11060-019-03220-0
Jermakowicz WJ, Mahavadi AK, Cajigas I, Dan L, Guerra S, Farooq G et al. Predictive modeling of brain tumor laser ablation dynamics. Journal of neuro-oncology. 2019 Jan 1. https://doi.org/10.1007/s11060-019-03220-0
Jermakowicz, Walter J. ; Mahavadi, Anil K. ; Cajigas, Iahn ; Dan, Lia ; Guerra, Santiago ; Farooq, Ghulam ; Shah, Ashish H. ; D’Haese, Pierre F. ; Ivan, Michael E. ; Jagid, Jonathan R. ; Komotar, Ricardo J. / Predictive modeling of brain tumor laser ablation dynamics. In: Journal of neuro-oncology. 2019.
@article{918306ae463543cdbe9ff4754c3e0ab7,
title = "Predictive modeling of brain tumor laser ablation dynamics",
abstract = "Introduction: Laser interstitial thermal therapy (LITT) is a novel MR thermometry-guided thermoablative tool revolutionizing the clinical management of brain tumors. A limitation of LITT is our inability to estimate a priori how tissues will respond to thermal energy, which hinders treatment planning and delivery. The aim of this study was to determine whether brain tumor LITT ablation dynamics may be predicted by features of the preoperative MRI and the relevance of these data, if any, to the recurrence of metastases after LITT. Methods: Intraoperative thermal damage estimate (TDE) map pixels representative of irreversible damage were retrospectively quantified relative to ablation onset for 101 LITT procedures. Raw TDE pixel counts and TDE pixel counts modelled with first order dynamics were related to eleven independent variables derived from the preoperative MRI, demographics, laser settings, and tumor pathology. Stepwise regression analysis generated predictive models of LITT dynamics, and leave-one-out cross validation evaluated the accuracy of these models at predicting TDE pixel counts solely from the independent variables. Using a deformable atlas, TDE maps were co-registered to the immediate post-ablation MRI, allowing comparison of predicted and actual ablation sizes. Results: Brain tumor TDE pixel counts modelled with first order dynamics, but not raw pixel counts, are correlated with the independent variables. Independent variables showing strong relations to the TDE pixel measures include T1 gadolinium and T2 signal, perfusion, and laser power. Associations with tissue histopathology are minimal. Leave-one-out analysis demonstrates that predictive models using these independent variables account for 77{\%} of the variance observed in TDE pixel counts. Analysis of metastases treated revealed a trend towards the over-estimation of LITT effects by TDE maps during rapid ablations, which was associated with tumor recurrence. Conclusions: Features of the preoperative MRI are predictive of LITT ablation dynamics and could eventually be used to improve the clinical efficacy with which LITT is delivered to brain tumors.",
keywords = "Coagulation, Dynamics, Perfusion imaging, Predictive modelling, Thermoablation",
author = "Jermakowicz, {Walter J.} and Mahavadi, {Anil K.} and Iahn Cajigas and Lia Dan and Santiago Guerra and Ghulam Farooq and Shah, {Ashish H.} and D’Haese, {Pierre F.} and Ivan, {Michael E.} and Jagid, {Jonathan R.} and Komotar, {Ricardo J.}",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/s11060-019-03220-0",
language = "English (US)",
journal = "Journal of Neuro-Oncology",
issn = "0167-594X",
publisher = "Kluwer Academic Publishers",

}

TY - JOUR

T1 - Predictive modeling of brain tumor laser ablation dynamics

AU - Jermakowicz, Walter J.

AU - Mahavadi, Anil K.

AU - Cajigas, Iahn

AU - Dan, Lia

AU - Guerra, Santiago

AU - Farooq, Ghulam

AU - Shah, Ashish H.

AU - D’Haese, Pierre F.

AU - Ivan, Michael E.

AU - Jagid, Jonathan R.

AU - Komotar, Ricardo J.

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Introduction: Laser interstitial thermal therapy (LITT) is a novel MR thermometry-guided thermoablative tool revolutionizing the clinical management of brain tumors. A limitation of LITT is our inability to estimate a priori how tissues will respond to thermal energy, which hinders treatment planning and delivery. The aim of this study was to determine whether brain tumor LITT ablation dynamics may be predicted by features of the preoperative MRI and the relevance of these data, if any, to the recurrence of metastases after LITT. Methods: Intraoperative thermal damage estimate (TDE) map pixels representative of irreversible damage were retrospectively quantified relative to ablation onset for 101 LITT procedures. Raw TDE pixel counts and TDE pixel counts modelled with first order dynamics were related to eleven independent variables derived from the preoperative MRI, demographics, laser settings, and tumor pathology. Stepwise regression analysis generated predictive models of LITT dynamics, and leave-one-out cross validation evaluated the accuracy of these models at predicting TDE pixel counts solely from the independent variables. Using a deformable atlas, TDE maps were co-registered to the immediate post-ablation MRI, allowing comparison of predicted and actual ablation sizes. Results: Brain tumor TDE pixel counts modelled with first order dynamics, but not raw pixel counts, are correlated with the independent variables. Independent variables showing strong relations to the TDE pixel measures include T1 gadolinium and T2 signal, perfusion, and laser power. Associations with tissue histopathology are minimal. Leave-one-out analysis demonstrates that predictive models using these independent variables account for 77% of the variance observed in TDE pixel counts. Analysis of metastases treated revealed a trend towards the over-estimation of LITT effects by TDE maps during rapid ablations, which was associated with tumor recurrence. Conclusions: Features of the preoperative MRI are predictive of LITT ablation dynamics and could eventually be used to improve the clinical efficacy with which LITT is delivered to brain tumors.

AB - Introduction: Laser interstitial thermal therapy (LITT) is a novel MR thermometry-guided thermoablative tool revolutionizing the clinical management of brain tumors. A limitation of LITT is our inability to estimate a priori how tissues will respond to thermal energy, which hinders treatment planning and delivery. The aim of this study was to determine whether brain tumor LITT ablation dynamics may be predicted by features of the preoperative MRI and the relevance of these data, if any, to the recurrence of metastases after LITT. Methods: Intraoperative thermal damage estimate (TDE) map pixels representative of irreversible damage were retrospectively quantified relative to ablation onset for 101 LITT procedures. Raw TDE pixel counts and TDE pixel counts modelled with first order dynamics were related to eleven independent variables derived from the preoperative MRI, demographics, laser settings, and tumor pathology. Stepwise regression analysis generated predictive models of LITT dynamics, and leave-one-out cross validation evaluated the accuracy of these models at predicting TDE pixel counts solely from the independent variables. Using a deformable atlas, TDE maps were co-registered to the immediate post-ablation MRI, allowing comparison of predicted and actual ablation sizes. Results: Brain tumor TDE pixel counts modelled with first order dynamics, but not raw pixel counts, are correlated with the independent variables. Independent variables showing strong relations to the TDE pixel measures include T1 gadolinium and T2 signal, perfusion, and laser power. Associations with tissue histopathology are minimal. Leave-one-out analysis demonstrates that predictive models using these independent variables account for 77% of the variance observed in TDE pixel counts. Analysis of metastases treated revealed a trend towards the over-estimation of LITT effects by TDE maps during rapid ablations, which was associated with tumor recurrence. Conclusions: Features of the preoperative MRI are predictive of LITT ablation dynamics and could eventually be used to improve the clinical efficacy with which LITT is delivered to brain tumors.

KW - Coagulation

KW - Dynamics

KW - Perfusion imaging

KW - Predictive modelling

KW - Thermoablation

UR - http://www.scopus.com/inward/record.url?scp=85068142113&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068142113&partnerID=8YFLogxK

U2 - 10.1007/s11060-019-03220-0

DO - 10.1007/s11060-019-03220-0

M3 - Article

C2 - 31240526

AN - SCOPUS:85068142113

JO - Journal of Neuro-Oncology

JF - Journal of Neuro-Oncology

SN - 0167-594X

ER -