Using prediction models to evaluate magnetic resonance image guided radiation therapy plans

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Online adaptive radiation therapy (ART) requires robust plan quality assessment.
• Magnetic resonance imaging (MRI)-guided linear accelerators (linacs) offer real-time adaptation capabilities.
• Assessing the quality and safety of dynamically adapted treatment plans is crucial for clinical adoption.

Purpose of the Study:

• To evaluate the quality and safety of daily online adaptive plans generated during MRI-guided radiation therapy.
• To compare the quality of online adapted plans with fully optimized offline plans.
• To assess the performance of artificial neural network (ANN) models in analyzing online adaptive plans.

Main Methods:

• Developed ANN models using offline treatment plans created post-simulation.
• Utilized these ANNs to analyze and compare online adaptive plans reoptimized in real-time.
• Compared plans adapted for 60Co-based radiotherapy versus MRI-linac systems.

Main Results:

• Approximately one-third of 60Co-adapted plans showed inferior quality compared to offline plans.
• MRI-linac adapted plans demonstrated quality essentially equivalent to offline plans.
• ANN models provided clear evidence of MRI-linac plan superiority over 60Co plans in most cases.

Conclusions:

• Online adaptive plans in MRI-guided radiation therapy can achieve quality comparable to offline plans.
• ANN models are effective tools for assessing the quality of online adaptive radiotherapy.
• MRI-linac systems show significant advantages in delivering high-quality adaptive radiation therapy compared to 60Co systems.