Time series analysis of dosimetric changes in target volumes and organs at risk monitored by cone beam computed tomography during radiotherapy for non-small-cell lung cancer

Background and Purpose

Technological advancements in radiotherapy have enhanced the ability to increase tumor doses while sparing normal tissues. We performed a time-series cluster analysis to characterize dynamic variations in dosimetry during radiotherapy.

Materials and Methods

We analyzed dosimetric variations of 40 patients with non-small-cell lung cancer who received 60 Gy over 30 daily cone beam computed tomography-guided fractions. Percentage deviations from planned doses across all fractions were calculated for organs at risk (OARs) (lungs, heart, spinal cord, and esophagus), gross tumor volume, clinical target volume (CTV), and planning target volume. K-means clustering with dynamic time-warping distances was applied to identify temporal dose patterns. The target volume with the lowest variance (95 % dose coverage of the CTV [CTV_D95]) was selected as a reference. OAR dose deviations were clustered using the elbow method. Mean temporal trends were plotted with 95 % confidence intervals.

Results

The CTV_D95 remained stable across all fractions and all patients and was used as a reference for time-series clustering. Cluster 1 included the majority of patients and exhibited low variability and predictable dose trajectories with gradual increases in the lung and esophagus doses and stable patterns in the heart and spinal cord. Cluster 2 showed high inter-fractional variability with progressively increasing OAR dose deviations.

Conclusions

Time-series clustering enabled early identification of patients with distinct dose evolution patterns. The stable trajectories of Cluster 1 may serve as internal references for adaptive radiotherapy. Variability in Cluster 2 underscores the need for routine dose monitoring to support timely offline adaptation.