Finding the optimal statistical model to describe target motion during radiotherapy delivery--a Bayesian approach

Area of Science:

• Medical Physics
• Radiation Oncology
• Biostatistics

Background:

• Early radiotherapy models assumed constant random errors in target displacement across patients.
• Recent advancements allow for patient-specific random error modeling, but rigorous comparison was previously unfeasible due to data limitations.

Purpose of the Study:

• To compare the goodness of fit for five different models of random error in target displacement.
• To identify the optimal statistical model for characterizing inter-patient variability in treatment errors.

Main Methods:

• Utilized real-world displacement data from 365 prostate cancer patients undergoing radical radiotherapy.
• Employed Bayesian statistics and Markov Chain Monte Carlo simulations to compare model goodness of fit.
• Assessed models ranging from constant random errors to various distributions of patient-specific errors.

Main Results:

• The model assuming inverse gamma distributed random errors demonstrated a significantly superior fit compared to all other evaluated models.
• This inverse gamma model effectively captures the variability in target displacement across patients.

Conclusions:

• Modeling random error as inverse gamma distributed is the most accurate approach for characterizing target displacement in radiotherapy.
• This improved modeling facilitates the development of more precise margin recipes and patient-specific correction strategies in radiation oncology.