Applications of Machine Learning to the High-Dose-Rate Cervical Brachytherapy Workflow: Applicator Prediction and Late Toxicity Modelling
| dc.contributor.advisor | McGeachy, Philip | |
| dc.contributor.advisor | Roumeliotis, Michael | |
| dc.contributor.author | Stenhouse, Kailyn | |
| dc.contributor.committeemember | Yanushkevich, Svetlana | |
| dc.contributor.committeemember | Wilms, Matthias | |
| dc.contributor.committeemember | Rink, Alexandra | |
| dc.contributor.committeemember | Roumeliotis, Michael | |
| dc.date | 2024-11 | |
| dc.date.accessioned | 2024-06-25T18:00:51Z | |
| dc.date.available | 2024-06-25T18:00:51Z | |
| dc.date.issued | 2024-06-20 | |
| dc.description.abstract | Cervical cancer is a significant global health issue, with brachytherapy being crucial for treating locally advanced disease. Brachytherapy involves inserting an applicator through the vaginal cavity to escalate radiation dose to the affected areas. Applicator geometries vary and impact the achievable dose distribution. Limited criteria exist to guide applicator selection, making it dependent on physician experience, presenting challenges in selection consistency and outcome variability. Suboptimal treatment may increase the likelihood of adverse effects post-treatment that impact quality of life. Additionally, there are currently limited comprehensive predictive models for treatment toxicities based on multi-institutional datasets. Machine learning can identify complex relationships and generate predictive models, and thus this thesis explored it’s potential to enhance the cervical cancer brachytherapy workflow by developing tools that assist physicians in making more informed treatment decisions. We first developed a decision-support tool for selecting the optimal treatment applicator, including hybrid interstitial needle arrangement. Using algorithm comparison and analyzing feature importance on retrospective data, we identified that boosted tree-based models combined with geometric features of the target volume provided the highest predictive accuracy. These models, validated through a prospective study, demonstrated comparable accuracy to expert radiation oncologists, with accuracies of 70% for applicator prediction and 82.5% for hybrid interstitial needle prediction. Machine learning predictions improved organ-at-risk dose compared to clinical predictions, demonstrating potential dosimetric benefit. Using a robust, multi-institutional dataset, we developed a Bayesian network approach to model late treatment toxicities, aiding in personalized and adaptive treatment strategies. We first developed a customized simulated annealing framework for optimizing network structures, integrating expert knowledge to ensure generated models have a logical structure that represent current clinical understanding. This framework demonstrated predictive performance comparable to out-of-box optimization methods, while providing a highly interpretable network structure. We explored potential clinical applications of these networks, including risk stratification, risk factor identification, and centre bias analysis. This thesis highlights novel applications of machine learning in supporting key aspects of the brachytherapy workflow for cervical cancer, potentially enhancing treatment quality and consistency, reducing treatment errors, and providing powerful clinical decision-support tools to physicians. | |
| dc.identifier.citation | Stenhouse, K. (2024). Applications of machine learning to the high-dose-rate cervical brachytherapy workflow: applicator prediction and late toxicity modelling (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. | |
| dc.identifier.uri | https://hdl.handle.net/1880/119021 | |
| dc.language.iso | en | |
| dc.publisher.faculty | Science | |
| dc.publisher.institution | University of Calgary | |
| dc.rights | University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. | |
| dc.subject | gynecologic brachytherapy | |
| dc.subject | medical physics | |
| dc.subject | artificial intelligence | |
| dc.subject | machine learning | |
| dc.subject | toxicity modelling | |
| dc.subject.classification | Artificial Intelligence | |
| dc.subject.classification | Physics--Radiation | |
| dc.subject.classification | Biophysics--Medical | |
| dc.title | Applications of Machine Learning to the High-Dose-Rate Cervical Brachytherapy Workflow: Applicator Prediction and Late Toxicity Modelling | |
| dc.type | doctoral thesis | |
| thesis.degree.discipline | Physics & Astronomy | |
| thesis.degree.grantor | University of Calgary | |
| thesis.degree.name | Doctor of Philosophy (PhD) | |
| ucalgary.thesis.accesssetbystudent | I require a thesis withhold – I need to delay the release of my thesis due to a patent application, and other reasons outlined in the link above. I have/will need to submit a thesis withhold application. |