Automatic IMRT planning via static field fluence prediction (AIP-SFFP): a deep learning algorithm for real-time...

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Intensity-modulated radiotherapy (IMRT) is a cornerstone of prostate cancer treatment.
• Manual IMRT planning is time-consuming and requires significant expertise.
• Automated planning solutions are needed to improve efficiency and consistency.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) algorithm, AIP-SFFP, for automated prostate IMRT planning.
• To assess the real-time planning efficiency and plan quality of the developed algorithm.
• To compare AIP-SFFP generated plans against knowledge-based planning (KBP) and clinical practice.

Main Methods:

• A custom DL neural network was developed for static field fluence prediction.
• Patient anatomy was represented by eight 2D projection images as input for AI training.
• AIP-SFFP bypasses inverse planning, directly generating fluence maps for dose calculation in a commercial system.

Main Results:

• All 14 independent test plans generated by AIP-SFFP met institutional criteria.
• Plan quality, including target coverage and organ-at-risk sparing, was comparable to KBP and clinical plans.
• AIP-SFFP generated each plan in under 20 seconds, demonstrating real-time efficiency.

Conclusions:

• AIP-SFFP is a successfully developed DL algorithm for automated prostate IMRT planning.
• The algorithm offers comparable plan quality and significant real-time efficiency.
• AIP-SFFP shows promise for immediate clinical application in radiation oncology.