Quantum Inspired Machine Learning Algorithms for Adaptive Radiotherapy
Pakela, Julia
2021
Abstract
Adaptive radiotherapy (ART) refers to the modification of radiotherapy treatment plans in response to patient anatomical and physiological changes over the course of treatment and has been recognized as an important step towards maximizing the curative potential of radiation therapy through personalized medicine. This dissertation explores the novel application of quantum physics principles and deep machine learning techniques to address three challenges towards the clinical implementation of ART: (1) efficient calculation of optimal treatment parameters, (2) adaptation to geometrical changes over the treatment period while mitigating associated uncertainties, and (3) understanding the relationship between individual patient characteristics and clinical outcomes. Applications of quantum and machine learning modeling in other fields support the potential of this novel approach. For efficient optimization, we developed and tested a quantum-inspired, stochastic algorithm for intensity-modulated radiotherapy: quantum tunnel annealing (QTA). By modeling the likelihood probability of accepting a higher energy solution after a particle tunneling through a potential energy barrier, QTA features an additional degree of freedom not shared by traditional stochastic optimization methods such as simulated annealing (SA). QTA achieved convergence up to 46.6% (26.8%) faster than SA for beamlet weight optimization and direct aperture optimization respectively. The results of this study suggest that the additional degree of freedom provided by QTA can improve convergence rates and achieve a more efficient and, potentially, effective treatment planning process. For geometrical adaptation, we investigated the feasibility of predicting patient changes across a fractionated treatment schedule using two approaches. The first was based on a joint framework (referred to as QRNN) employing quantum mechanics in combination with deep recurrent neural networks (RNNs). The second approach was developed based on a classical framework (MRNN), which modelled patient anatomical changes as a Markov process. We evaluated and compared these two approaches’ performance characteristics using a dataset of 125 head and neck cancer patients who received fractionated radiotherapy. The MRNN framework exhibited slightly better performance than the QRNN framework, with MRNN(QRNN) validation area under the receiver operating characteristic curve (AUC) scores [95% CI] of 0.742 [0.721-0.763] (0.675 [0.64-0.71]), 0.709 [0.683-0.735] (0.656 [0.634-0.677]), 0.724 [0.688-0.76] (0.652 [0.608-0.696]), and 0.698 [0.682-0.714] (0.605 [0.57-0.64]) for system state vector sizes of 4, 6, 8, and 10, respectively. A similar trend was also observed when the fully trained models were applied to an external testing dataset of 20 patients. These results suggest that these stochastic models provide added value in predicting patient changes during the course of adaptive radiotherapy. Towards understanding the relationship between patient characteristics and clinical outcomes, we performed a series of studies which investigated the use of quantitative patient features for predicting clinical outcomes in laryngeal cancer patients who underwent treatment in a bioselection paradigm based on surgeon-assessed response to induction chemotherapy. Among the features investigated from CT scans taken before and after induction chemotherapy, two (gross tumor volume change between pre- and post-induction chemotherapy, and nodal stage) had prognostic value for predicting patient outcomes using standard regression models. Artificial neural networks did not improve predictive performance in this case. Taken together, the significance of these studies lies in their contribution to the body of knowledge of medical physics and in their demonstration of the use of novel techniques which incorporate quantum mechanics and machine learning as a joint framework for treatment planning optimization and prediction of anatomical patient changes over time.Deep Blue DOI
Subjects
Machine Learning Adaptive Radiotherapy Quantum Algorithms
Types
Thesis
Metadata
Show full item recordCollections
Remediation of Harmful Language
The University of Michigan Library aims to describe library materials in a way that respects the people and communities who create, use, and are represented in our collections. Report harmful or offensive language in catalog records, finding aids, or elsewhere in our collections anonymously through our metadata feedback form. More information at Remediation of Harmful Language.
Accessibility
If you are unable to use this file in its current format, please select the Contact Us link and we can modify it to make it more accessible to you.