Development of geometrical parameters to describe anatomical changes and predict the need for radiotherapy replanning in head and neck cancer patients

Head and neck cancer patients undergoing radiotherapy may experience significant anatomical changes due to weight loss and tumor shrinkage. These changes can impact the effectiveness of the initial treatment plan, potentially necessitating treatment replanning. However, ad hoc replanning requires additional clinical staff time, which can lead to suboptimal and stressful treatment planning. This interruption in workflow can impact many patients. Furthermore, there is currently no established method for determining the total amount of anatomical variation in the head and neck region to decide whether replanning is necessary.This thesis project aimed to identify and create metrics based on patient anatomical structures, that can describe the anatomical alterations that patients may experience over the course of the treatment, and influence decisions regarding treatment replanning. These parameters were used in the development of a machine learning classification model to predict if and when patients should undergo replanning. This model accounts for the progression of the information over time, which will be achieved with the creation of an automatic extraction pipeline.This study included 120 head and neck cancer patients treated at the McGill University Health Centre. Based on the 3D shape and 2D contours of radiotherapy structures, we defined 43 parameters, such as the average or Chamfer distance, to describe body shrinkage. We calculated the rate of change of these parameters across fractions through linear regression analysis and obtained the variation of each parameter with respect to initial values both analyses provided significant insights for evaluating replanning. The Mann-Whitney U test and the Bonferroni correction were used for statistical analysis, which revealed significant differences between replanned and non-replanned patients as early as fraction 4 in the case of rate of changes, indicating the potential for early prediction. Specific fraction machine learning models for fractions 5, 10, and 15 were built using the parameters, clinical data, and feature selection techniques. To estimate the performance of the models, the repeated stratified 5-fold cross-validation resampling technique was used with the Area Under Curve (AUC) as a performance metric. The best multivariate models for fractions 5, 10, and 15 yielded a training AUC score of 0.86 [0.65 - 1.00], 0.87 [0.71 - 1.00], and 0.92 [0.78 - 1.00], respectively. When testing the models in a hold-out set of patients, each showed AUC values of 0.82, 0.83, and 0.79. The developed machine learning models based on information from radiotherapy structures demonstrate the ability of early identification of patients who may need replanning, which if implemented, will enhance patient outcomes, streamline clinical workflows, and reduce costs in radiotherapy treatments. However, further analysis needs to be done to overcome false positive and negative cases

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.