Machine Learning Framework to Reduce Patient-Reported Dysphagia in Head and Neck Radiotherapy
Date
2024-06-13
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Radiotherapy treatment has become more effective in the 21st century due to advancements in treatment technologies such as image-guided radiotherapy and volumetric modulated arc therapy. Machine learning algorithms can provide advanced analytics based on the clinical, imaging, and dosimetric data available through radiotherapy treatments. Head and neck tumours are a site of specific interest, as these tumours often lie near many organs at risk which may cause radiation-induced acute or late toxicities while damaging the tumour. Among these organs are the muscles involved in the swallowing process that may cause dysphagia, or difficulty swallowing, after treatment. In this work, we investigated factors correlated with late dysphagia evaluated using patient-reported outcomes, with the goal of reducing this significant toxicity after head and neck radiotherapy. We examined anatomical changes to the pharyngeal constrictor muscles hypothesized to be correlated with dysphagia or cause significant dosimetric changes outside the random error associated with radiotherapy treatments. We found a significant increase in pharyngeal constrictors thickness and identified high dose gradients were closer to the pharyngeal constrictors in patients with dysphagia. Both features may be useful for identifying patients at risk of late patient-reported dysphagia. We then examined treatment planning dose constraints used to develop radiotherapy treatment plans. The mean dose constraints currently used within the literature did not classify patients well into the patient-reported symptom groups. Metrics were identified for the pharyngeal constrictor muscles and their substructures. The results suggest additional dose constraints to the pharyngeal constrictor muscles could reduce late patient-reported dysphagia. The final study examined advanced imaging features to develop predictive machine learning models for late patient-reported dysphagia. Current models used by speech language pathologists are limited to acute models. We successfully created models to predict late patient-reported dysphagia with sufficient accuracy and sensitivity, which may improve the identification and follow-up of patients at risk. Through this work, we were able to create recommendations based on several forms of intervention: adaptive planning, treatment planning, and predictive modelling. Further work, including prospective studies, may be required prior to clinical implementation of treatment planning or predictive modelling recommendations.
Description
Keywords
dysphagia, radiotherapy, head and neck, machine learning
Citation
Paetkau, D. O. (2024). Machine learning framework to reduce patient-reported dysphagia in head and neck radiotherapy (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.