Training and testing of machine learning techniques to aid in prediction of patients requiring adaptive treatments for head and neck radiotherapy.

Abstract

Adaptive radiotherapy (ART) offers a tailored approach to radiotherapy treatment and has been shown to be beneficial to patients undergoing treatment for head and neck carcinoma. The challenge lies in prospectively identifying patients who will benefit from ART intervention at the planning stage. This study presents the assessment of AI-based predictive models aimed to address this challenge. Retrospective data from 100 head and neck patients were analysed, encompassing various patient features, including weight, neck dimensions, body volume, and target volumes. The training phase began with a decision tree algorithm, which was compared to a selection of other suitable classifiers, being: random forest, bagging, adaBoost and gradient boosting. Model performance was assessed using accuracy, F1 score, and cross-validation accuracy. Initial features in the classifier were selected based on expert (RO) opinion; feature selection was done to refine the final model. The final model was tested on new patient data (N = 110). Final performance was assessed using precision, recall, specificity, and sensitivity. The initial model exhibited F1 score 65%, test accuracy 60%, and cross-validation accuracy 72%. However, when tested on new data, a notable prevalence of false positives (21 cases) was observed. Analysis of these cases revealed a spectrum of adaptive interventions leading to reclassification of these instances, indicating the model's ability to discern patients requiring varying levels of intervention at the local centre. The Random Forest Decision Tree demonstrates promise in identifying head and neck carcinoma patients who are likely to require ART. The high number of false positives, initially perceived as inaccuracies, underscores the model's ability to detect patients in need of ART, even when a complete rescan is not warranted. This offers the potential to shift from a reactive to a proactive approach to ART. The ML-based predictive model offers a nuanced approach to patient selection, ensuring those who require ART, in varying degrees, are identified and treated accordingly. The transition from a reactive to a proactive approach has potential to improve patient outcomes and streamline clinical practice in ART.