A deep learning framework for radiotherapy dose prediction: from MLC motion to patient-specific dose perturbation

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Modern radiotherapy relies on multi leaf collimators (MLCs) for precise dose sculpting.
• Patient-specific quality assurance (PSQA) traditionally involves passive verification of treatment delivery.
• Accurate modeling of MLC-related delivery uncertainties is crucial for enhancing radiotherapy precision.

Purpose of the Study:

• To develop a physics-informed, interpretable framework for modeling MLC delivery uncertainties.
• To predict the dosimetric impact of MLC deviations on radiotherapy dose distributions.
• To transition PSQA from passive verification to proactive intervention by quantifying delivery uncertainty.

Main Methods:

• Analysis of trajectory log files from 1954 volumetric modulated arc therapy (VMAT) plans across two MLC designs.
• Modeling of control point level MLC and gantry deviations using motion metrics and a novel temporal latency parameter (ΔtMLC).
• Projection of deviations into 3D anatomy to create an MLC Position Deviation Texture (MPDT) and input into a dual pathway U-Net for dose prediction.

Main Results:

• MLC deviation patterns were design-specific with strong linear/quadratic relationships (average R² = 0.989).
• The MPDT-enhanced framework achieved superior dose prediction accuracy (SSIM = 0.996, PMAE = 0.39%) compared to conventional models.
• Improved consistency in gamma passing rate (GPR) was observed (ΔGPR = 2.40% vs. 3.92%), persisting in clinical testing.

Conclusions:

• The study presents a scalable paradigm for PSQA by integrating interpretable mechanical modeling with deep learning.
• This framework quantifies radiotherapy delivery uncertainty and traces dose deviations to their physical origins.
• The developed approach enables proactive intervention, enhancing the safety and efficacy of precision oncology treatments.