(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: Modern radiotherapy stands to benefit from the ability to efficiently adapt plans mid-treatment in response to non-random geometric variations such as those caused by tumor shrinkage. A promising strategy is to develop a robust framework which, given an initial state defined by relevant patient-attributes, can predict future states based on pre-learned patterns from a well-defined patient population—reminiscent of the time-evolution of a stationary state in quantum mechanics. Here, we present a proof-of-concept study, which investigates the feasibility of predicting patient changes across a fractionated treatment schedule using a joint framework employing quantum mechanics in combination with deep recurrent neural networks (RNNs).
Methods: Data were obtained from 92 head and neck cancer patients who received fractionated radiotherapy at one institution. Clinical target volume (CTV) structure volumes were calculated from daily CBCT images and used to train an RNN to predict the fraction sequence’s next step. The differences between the testing dataset’s predicted anatomical volumes and the ground truth volumes were used to assess the feasibility of deriving a Hamiltonian matrix which predicts the time course of non-random changes in a patient’s volume.
Results: CBCT volumes mapped into discrete state values using Lloyd-Max quantization (4 levels) and encoded as one-hot binary vectors, which were used for RNN training. The results were evaluated using 5-fold cross-validation. The average AUC during validation was 0.689±0.037. This information will be used to derive the Hamiltonian of the time independent Schrodinger equation.
Conclusion: This study investigated the feasibility of a novel framework for predicting changes in patient geometry over time by combining quantum mechanics with RNN techniques to improve robustness. Our results indicate that predictive information can be learned from sequential patient data mapped to discrete unitary states, representing a fundamental first step towards deriving a Hamiltonian which predicts temporal changes in a patient’s initial state.
Funding Support, Disclosures, and Conflict of Interest: Research reported in this abstract was supported by the National Cancer Institute of the National Institutes of Health under award number 1R01CA233487-01A1. We have no additional conflicts of interest to report.
Keywords
Cone-beam CT, Modeling, Radiation Therapy
Taxonomy
TH- Dataset Analysis/Biomathematics: Machine learning techniques
Contact Email
