MO‐A‐137‐06: A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning

T. Long, M. Matuszak, M. Schipper, M. Epelman, F. Kong, R. Ten Haken, E. Romeijn

Research output: Contribution to journalArticle

Abstract

Purpose: We have demonstrated that the ratio between a lung cancer patient's TGFβ1 level two weeks into treatment versus pre‐treatment is a predictive biomarker for radiation induced lung toxicity (RILT). Instead of adapting to this new information only when the data is available, we present a stochastic optimization model that can explicitly incorporate this future knowledge into adaptive lung radiation therapy treatment planning. Methods: A two‐stage stochastic treatment plan optimization model is developed that accommodates knowledge of a patient's predisposition to RILT that would be obtained during treatment. The goal of the stochastic model is to maximize the population probability of local tumor control (LTC) after 2 years while controlling the probability of RILT. Two adaptive strategies are considered: (1) bounding the expected proportion of the entire population that will experience RILT, and (2) bounding the probability that each individual patient will experience RILT. These strategies were compared to a non‐adaptive treatment planning strategy that uses a simple bound on the mean lung dose to control the probability of RILT. Results: This technique is applied to lung cancer cases that are treated in daily fractions over a six‐week period. The tradeoffs between the probability of RILT and LTC are observed for each strategy. Both adaptive strategies dominate (are always better than) the non‐adaptive treatment plans in the worst‐case sense (i.e., highest probability among all treated patients), while adaptive strategy (1) dominates it in the population sense in terms of the tradeoff between probability of RILT and expected probability of LTC. Conclusion: By incorporating future knowledge into a two‐stage stochastic IMRT treatment planning model, superior adaptive treatment plans can be obtained that improve both population and worst‐case outcomes for patients. In addition, this modeling paradigm provides treatment planners a tool to help personalize treatment plans for individual patients. Troy Long is on a NSF Graduate Research Fellowship. Data collection was assisted by R01CA142840 grant.

Original languageEnglish (US)
Pages (from-to)388
Number of pages1
JournalMedical Physics
Volume40
Issue number6
DOIs
StatePublished - 2013
Externally publishedYes

Fingerprint

Radiotherapy
Lung
Radiation
Therapeutics
Population
Lung Neoplasms
Neoplasms
Organized Financing
Biomarkers
Research

ASJC Scopus subject areas

  • Biophysics
  • Radiology Nuclear Medicine and imaging

Cite this

Long, T., Matuszak, M., Schipper, M., Epelman, M., Kong, F., Ten Haken, R., & Romeijn, E. (2013). MO‐A‐137‐06: A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning. Medical Physics, 40(6), 388. https://doi.org/10.1118/1.4815209

MO‐A‐137‐06 : A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning. / Long, T.; Matuszak, M.; Schipper, M.; Epelman, M.; Kong, F.; Ten Haken, R.; Romeijn, E.

In: Medical Physics, Vol. 40, No. 6, 2013, p. 388.

Research output: Contribution to journalArticle

Long, T, Matuszak, M, Schipper, M, Epelman, M, Kong, F, Ten Haken, R & Romeijn, E 2013, 'MO‐A‐137‐06: A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning', Medical Physics, vol. 40, no. 6, pp. 388. https://doi.org/10.1118/1.4815209
Long, T. ; Matuszak, M. ; Schipper, M. ; Epelman, M. ; Kong, F. ; Ten Haken, R. ; Romeijn, E. / MO‐A‐137‐06 : A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning. In: Medical Physics. 2013 ; Vol. 40, No. 6. pp. 388.
@article{5648be064dc54968ab0bd3d07970dc3d,
title = "MO‐A‐137‐06: A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning",
abstract = "Purpose: We have demonstrated that the ratio between a lung cancer patient's TGFβ1 level two weeks into treatment versus pre‐treatment is a predictive biomarker for radiation induced lung toxicity (RILT). Instead of adapting to this new information only when the data is available, we present a stochastic optimization model that can explicitly incorporate this future knowledge into adaptive lung radiation therapy treatment planning. Methods: A two‐stage stochastic treatment plan optimization model is developed that accommodates knowledge of a patient's predisposition to RILT that would be obtained during treatment. The goal of the stochastic model is to maximize the population probability of local tumor control (LTC) after 2 years while controlling the probability of RILT. Two adaptive strategies are considered: (1) bounding the expected proportion of the entire population that will experience RILT, and (2) bounding the probability that each individual patient will experience RILT. These strategies were compared to a non‐adaptive treatment planning strategy that uses a simple bound on the mean lung dose to control the probability of RILT. Results: This technique is applied to lung cancer cases that are treated in daily fractions over a six‐week period. The tradeoffs between the probability of RILT and LTC are observed for each strategy. Both adaptive strategies dominate (are always better than) the non‐adaptive treatment plans in the worst‐case sense (i.e., highest probability among all treated patients), while adaptive strategy (1) dominates it in the population sense in terms of the tradeoff between probability of RILT and expected probability of LTC. Conclusion: By incorporating future knowledge into a two‐stage stochastic IMRT treatment planning model, superior adaptive treatment plans can be obtained that improve both population and worst‐case outcomes for patients. In addition, this modeling paradigm provides treatment planners a tool to help personalize treatment plans for individual patients. Troy Long is on a NSF Graduate Research Fellowship. Data collection was assisted by R01CA142840 grant.",
author = "T. Long and M. Matuszak and M. Schipper and M. Epelman and F. Kong and {Ten Haken}, R. and E. Romeijn",
year = "2013",
doi = "10.1118/1.4815209",
language = "English (US)",
volume = "40",
pages = "388",
journal = "Medical Physics",
issn = "0094-2405",
publisher = "AAPM - American Association of Physicists in Medicine",
number = "6",

}

TY - JOUR

T1 - MO‐A‐137‐06

T2 - A Stochastic Optimization Approach to Adaptive Lung Radiation Therapy Treatment Planning

AU - Long, T.

AU - Matuszak, M.

AU - Schipper, M.

AU - Epelman, M.

AU - Kong, F.

AU - Ten Haken, R.

AU - Romeijn, E.

PY - 2013

Y1 - 2013

N2 - Purpose: We have demonstrated that the ratio between a lung cancer patient's TGFβ1 level two weeks into treatment versus pre‐treatment is a predictive biomarker for radiation induced lung toxicity (RILT). Instead of adapting to this new information only when the data is available, we present a stochastic optimization model that can explicitly incorporate this future knowledge into adaptive lung radiation therapy treatment planning. Methods: A two‐stage stochastic treatment plan optimization model is developed that accommodates knowledge of a patient's predisposition to RILT that would be obtained during treatment. The goal of the stochastic model is to maximize the population probability of local tumor control (LTC) after 2 years while controlling the probability of RILT. Two adaptive strategies are considered: (1) bounding the expected proportion of the entire population that will experience RILT, and (2) bounding the probability that each individual patient will experience RILT. These strategies were compared to a non‐adaptive treatment planning strategy that uses a simple bound on the mean lung dose to control the probability of RILT. Results: This technique is applied to lung cancer cases that are treated in daily fractions over a six‐week period. The tradeoffs between the probability of RILT and LTC are observed for each strategy. Both adaptive strategies dominate (are always better than) the non‐adaptive treatment plans in the worst‐case sense (i.e., highest probability among all treated patients), while adaptive strategy (1) dominates it in the population sense in terms of the tradeoff between probability of RILT and expected probability of LTC. Conclusion: By incorporating future knowledge into a two‐stage stochastic IMRT treatment planning model, superior adaptive treatment plans can be obtained that improve both population and worst‐case outcomes for patients. In addition, this modeling paradigm provides treatment planners a tool to help personalize treatment plans for individual patients. Troy Long is on a NSF Graduate Research Fellowship. Data collection was assisted by R01CA142840 grant.

AB - Purpose: We have demonstrated that the ratio between a lung cancer patient's TGFβ1 level two weeks into treatment versus pre‐treatment is a predictive biomarker for radiation induced lung toxicity (RILT). Instead of adapting to this new information only when the data is available, we present a stochastic optimization model that can explicitly incorporate this future knowledge into adaptive lung radiation therapy treatment planning. Methods: A two‐stage stochastic treatment plan optimization model is developed that accommodates knowledge of a patient's predisposition to RILT that would be obtained during treatment. The goal of the stochastic model is to maximize the population probability of local tumor control (LTC) after 2 years while controlling the probability of RILT. Two adaptive strategies are considered: (1) bounding the expected proportion of the entire population that will experience RILT, and (2) bounding the probability that each individual patient will experience RILT. These strategies were compared to a non‐adaptive treatment planning strategy that uses a simple bound on the mean lung dose to control the probability of RILT. Results: This technique is applied to lung cancer cases that are treated in daily fractions over a six‐week period. The tradeoffs between the probability of RILT and LTC are observed for each strategy. Both adaptive strategies dominate (are always better than) the non‐adaptive treatment plans in the worst‐case sense (i.e., highest probability among all treated patients), while adaptive strategy (1) dominates it in the population sense in terms of the tradeoff between probability of RILT and expected probability of LTC. Conclusion: By incorporating future knowledge into a two‐stage stochastic IMRT treatment planning model, superior adaptive treatment plans can be obtained that improve both population and worst‐case outcomes for patients. In addition, this modeling paradigm provides treatment planners a tool to help personalize treatment plans for individual patients. Troy Long is on a NSF Graduate Research Fellowship. Data collection was assisted by R01CA142840 grant.

UR - http://www.scopus.com/inward/record.url?scp=85024820702&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85024820702&partnerID=8YFLogxK

U2 - 10.1118/1.4815209

DO - 10.1118/1.4815209

M3 - Article

AN - SCOPUS:85024820702

VL - 40

SP - 388

JO - Medical Physics

JF - Medical Physics

SN - 0094-2405

IS - 6

ER -