Author ORCID Identifier
0000-0003-2327-4226
Date of Graduation
8-2024
Document Type
Dissertation (PhD)
Program Affiliation
Medical Physics
Degree Name
Doctor of Philosophy (PhD)
Advisor/Committee Chair
Stephen Kry, Ph.D.
Committee Member
Rebecca Howell, Ph.D.
Committee Member
David Jaffray, Ph.D.
Committee Member
Julianne Pollard-Larkin, Ph.D.
Committee Member
Christine Peterson, Ph.D.
Committee Member
Laurence Court, Ph.D.
Abstract
IROC’s mission is to help ensure consistent and comparable, high-quality radiotherapy across clinics that participate in national clinical trials. To obtain this mission, IROC’s phantom program provides a third-party end-to-end check of the clinical workflow of a patient receiving radiotherapy. The goal of the phantom audit is to compare the dose delivered to the dose planned by the treatment system ensuring dose delivery accuracy. While IROC’s phantoms are better equipped to catch dose delivery errors compared to a clinic’s QA process, the end-to-end process and reporting of results is time-consuming creating a bottleneck for clinical trial participation. Furthermore, IROC’s passing criteria must remain loose to allow for sufficient powering of clinical trials in a timely manner and even if a clinic fails the phantom, meaningful and productive feedback is difficult to provide lowering the quality of radiotherapy allow within clinical trials affecting the overall results.
This aims of this work seek to remedy these two issues of IROC’s phantom program through machine learning, which, in turn, will improve clinical trials. This will be accomplished by 1) identifying and understanding the important factors that drive phantom failures across multiple treatment modalities through machine learning, and 2) development of a virtual phantom model for predicting dose delivery accuracy. Three of IROC’s phantoms: head and neck, stereotactic head, and thoracic, were retrospectively analyzed with random forest algorithm to predict phantom performance metrics that determine dose delivery accuracy. These three phantoms were chosen because they are the majority of those required by clinical trials for participation. For each phantom study, properties of the treatment were collected about the treatment system and calculated from the treatment plan. Furthermore, important factors for random forest algorithm were captured to highlight the underlying differences between passing and failing phantoms. With the head and neck phantom, a virtual phantom model was developed which expanded the metrics calculated from the treatment plan, compared different machine learning algorithms and feature selection schemes, and used interpretability algorithms to further understand the contributing factors between passing and failing a phantom.
This study will provide an avenue to shorten the time clinics can receive results from a phantom audit ensuring faster enrollment for patients to clinical trials and furthermore, feedback will be provided to clinics to raise the quality of their radiotherapy and improve their clinical workflow. Clinical trials will be positively impacted by reducing the overall time to accumulate sufficient power for meaningful results and due to the improvement of quality in radiotherapy, the noise and variation within the clinical trial will be reduced quickening even further the time it takes to complete clinical trials. This work has even further reach by improving current clinical workflows which will inevitably lead to increase patient safety and outcome.
Keywords
IROC, Phantoms, machine learning, radiotherapy