Related Experiment Videos
Uncertainties in model-based outcome predictions for treatment planning
J O Deasy1, K S Chao, J Markman
1Department of Radiation Oncology, Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, MO 63110, USA. deasy@radonc.wustl.eu
International Journal of Radiation Oncology, Biology, Physics
|December 1, 2001
Summary
This study introduces a method to quantify uncertainties in radiation therapy outcome predictions, enhancing treatment plan reliability. It provides visual tools to assess prediction confidence for individual patients and treatment plans.
Area of Science:
- Radiation Oncology
- Medical Physics
- Biostatistics
Background:
- Model-based outcome predictions in radiation therapy often lack uncertainty quantification.
- Reliable predictions are crucial for treatment planning, especially for normal tissue complication probability (NTCP) and salivary function.
Purpose of the Study:
- To develop and present a method for assessing the reliability of treatment-plan-specific outcome predictions.
- To incorporate uncertainties into predictions of normal tissue complications and functional outcomes.
Main Methods:
- A bootstrap-based approach is used to estimate parameter uncertainties from original data.
- The framework handles both continuous (e.g., salivary function) and dichotomous (e.g., NTCP) outcome predictions.
- Residual uncertainty is modeled by adding random components based on model fit.
Main Results:
- The method generates uncertainty histograms for visualizing prediction reliability for individual patients.
- Significant variability in prediction certainty was observed for salivary function in head-and-neck cancer patients.
- Uncertainty quantification improves accuracy for dichotomous endpoints (TCP, NTCP) and enables reliable ranking of competing treatment plans.
Conclusions:
- A comprehensive framework for incorporating uncertainties into outcome predictions is established.
- Uncertainty histograms offer a clear method for evaluating the reliability of treatment plan predictions.