Updating a clinical Knowledge-Based Planning prediction model for prostate radiotherapy.

Area of Science:

• Radiation Oncology
• Medical Physics
• Computational Biology

Background:

• Clinical knowledge-based planning (KBP) models for prostate radiotherapy require updates to maintain relevance.
• Changes in clinical practice necessitate adaptation of KBP models.

Purpose of the Study:

• To compare two distinct update approaches for KBP models through longitudinal analysis.
• To evaluate the impact of model updates on plan quality and robustness.

Main Methods:

• A KBP model for prostate therapy was updated four times using two strategies: maximizing library size (Mt) and maximizing mean sample quality (Rt).
• Plan quality was assessed using the Plan Quality Metric (PQM), and plan complexity was monitored.
• Outcomes were compared to the common ancestor model.

Main Results:

• Both update methods improved organ-at-risk (OAR) sparing by 3.9% to 19.2% compared to human-generated plans.
• Target coverage and homogeneity slightly decreased, while plan complexity remained largely unchanged.
• Increasing sample size enhanced prediction reliability and goodness-of-fit; increasing sample quality improved outcomes but reduced model reliability.

Conclusions:

• Repeated updates enhance KBP model robustness, reliability, and automated plan quality.
• Periodically expanding the model sample and removing low-quality plans optimizes update benefits and workload.