Photo Credit: Jacob Wackerhausen
The following is a summary of “Cluster-Based Toxicity Estimation of Osteoradionecrosis via Unsupervised Machine Learning: Moving Beyond Single Dose-Parameter Normal Tissue Complication Probability by Using Whole Dose-Volume Histograms for Cohort Risk Stratification,” published in the March 2024 issue of Oncology by Hosseinian et al.
The objective of this study was to address the limitations of existing models for estimating the normal tissue complication probability associated with osteoradionecrosis (ORN) of the mandible by introducing a more robust and clinically reliable approach. Leveraging unsupervised machine learning techniques specifically designed to capture the structural intricacies of the data, the study aimed to provide an enhanced model for evaluating ORN risk by incorporating the entire radiation dose distribution on the mandible.
The analysis was conducted on retrospective data encompassing 1,259 patients treated for head and neck cancer (HNC) at XXX between 2005 and 2015. Over a minimum 12-month post-therapy follow-up period, 173 patients within this cohort (13.7%) developed ORN ranging from grades I to IV. Utilizing the K-means clustering method, the study identified distinct clusters within these patients’ mandibular dose-volume histograms (DVHs). Subsequently, a soft-margin support vector machine (SVM) was employed to delineate cluster borders and partition the dose-volume space, facilitating the calculation of ORN risk for each dose-volume region based on incidence rates and other relevant clinical risk factors.
The analysis revealed six identifiable clusters among the DVHs, with the soft-margin SVM effectively partitioning the dose-volume space into distinct regions characterized by varying risk indices. The study delineated specific risk regions based on pre-radiation dental extraction status, a significant non-dose-related risk factor for ORN. Ultimately, the findings offer a comprehensive visual risk-assessment tool for ORN based on the entire DVH and pre-radiation dental extraction status, providing valuable insights for dose optimization strategies tailored to different risk levels. Overall, this study presents a sophisticated approach to ORN risk evaluation, leveraging unsupervised learning techniques on a large-scale dataset to enhance clinical decision-making in managing HNC patients.
Source: sciencedirect.com/science/article/pii/S0360301624003298