Automatic radiotherapy planning for deliverable plans using deep learning dose prediction and dose rings optimization...

Area of Science:

• Radiation Oncology
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Deep learning (DL) shows promise for automatic radiotherapy (RT) planning.
• Integrating DL-predicted dose distributions into mainstream treatment planning systems (TPSs) for clinically deliverable plans remains a challenge.

Purpose of the Study:

• To assess the feasibility and accuracy of an automated VMAT and IMRT planning approach.
• This method combines DL dose prediction with dose rings optimization for universally deliverable plans.

Main Methods:

• A 3D Fusion Residual Unet (F-ResUnet) DL network predicted dose distributions for 440 gynecological cancer (GC) patients.
• Predicted doses were discretized into dose rings for automatic optimization in two mainstream TPSs.
• Plan deliverability was confirmed via patient-specific quality assurance (PSQA).

Main Results:

• DL dose predictions were clinically acceptable, exceeding 95% target coverage.
• Automatic plans demonstrated comparable target coverage (over 96.5%) to clinical plans.
• Significant reductions in organ-at-risk (OAR) doses were observed, with high gamma passing rates (e.g., 99.1% for VMAT at 3%/3mm).

Conclusions:

• The combined DL dose prediction and dose rings optimization method is feasible.
• This approach successfully generates universally deliverable VMAT and IMRT plans for GC patients.
• The method offers a viable solution for integrating DL into clinical RT workflows.