Determining The Optimal Modality For The Treatment Of Brain Metastases

Jason Rockhill1, Phillips Mark1, Ghate Archis1, Kim Minsun1, Zhang Pengbo1

1University of Washington, Seattle, USA

Keywords: gamma knife, brain metastasis, radiosurgery, radiotherapy, resection

Interactive Manuscript

Ask Questions of this Manuscript:



     To use a Markov Decision Process (MDP) to determine the optimal treatment modality for brain metastases with the expectation that multiple treatments may be needed. Gamma Knife radiosurgery (GK-SRS) is emerging as an important treatment option for the treatment of brain metastases, with the other options being whole brain radiation (WBRT) and surgical excision (S). Early detection, the efficacy of these modalities, and improved management of the primary disease make it increasingly likely that patients will require subsequent treatment for new brain metastases and/or the recurrence of previously treated tumors.
     Given the morbidity and effectiveness of each of the treatment options, it is difficult to assess the best treatment policy.
     A Markov Decision Process was constructed and solved for a range of clinical conditions. AMarkov decision process is one means of dynamically optimizing treatment decisions under the multiple uncertainties regarding the likelihoods of the patient surviving, new metatstases emerging and the effectiveness of any given therapy. An MDP was constructed with four actions (GK-SRS, WBRT, S, and Wait). Each brain metastasis was characterized by its size and whether it was operable. The state space included from 1 – N mets (initially N was set to 3). Transition probabilities and rewards were obtained from the literature and expert opinion
     The model provided the optimal treatment policy throughout the course of the disease. Initial work has focussed on building the model and determining proper values of the probabilities and rewards. An important consideration is the overall size of the problem versus the time to solve. Comparisons between the predictions and current practice were used to verify the correctness of the model.
     This is a retrospective study.
     Preliminary work has demonstrated that a realistic MDP can be constructed for the problem of determining the optimal treatment modality. 
     The long term goal is to provide a method for selecting the best modality under the conditions that patients survive long enough that treatment of brain metastases may occur more than once. Future work will incorporate transition probabilities and outcomes from the UW treatment experience.


Project Roles:

J. Rockhill (), P. Mark (), G. Archis (), K. Minsun (), Z. Pengbo ()