Quantitative Evaluation of a Fully Automated Planning Solution for Prostate-Only and Whole-Pelvic Radiotherapy

Area of Science:

• Radiation Oncology
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Developing automated treatment planning systems for radiation oncology is crucial for efficiency and consistency.
• Prostate cancer treatment planning requires precise dose delivery to the target while sparing organs-at-risk (OARs).

Purpose of the Study:

• To evaluate an automated, end-to-end treatment planning pipeline for prostate-only and whole-pelvic radiation therapy.
• To assess the quality, clinical acceptability, and machine deliverability of AI-generated plans.

Main Methods:

• A deep learning model was developed for dose prediction and integrated with a direct aperture VMAT optimization.
• The pipeline was trained on 238 cases and validated on 86 cases, with a clinical evaluation on 40 additional cases.
• Plan quality was assessed using dose-volume histogram (DVH) metrics, and deliverability was verified using portal dosimetry (gamma index).

Main Results:

• Automated plans met dose constraints with minor deviations, showing quality comparable to manually generated plans.
• Portal dosimetry confirmed successful deliverability of the automated plans.
• No significant correlation was found between monitor units (MUs), modulation complexity scores (MCSv), and gamma criteria.

Conclusions:

• A deep learning-based automated pipeline is feasible for generating high-quality radiation therapy plans for prostate cancer.
• The automated plans are competitive with manual plans and clinically approved for dosimetry and machine deliverability.