Fluence-map generation for prostate intensity-modulated radiotherapy planning using a deep-neural-network

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Deep neural networks (DNNs) have shown promise in predicting radiotherapy dose distributions.
• Automating intensity-modulated radiotherapy (IMRT) plan generation is crucial for efficiency and quality.
• Current DNN applications often require inverse planning, adding complexity.

Purpose of the Study:

• To develop and evaluate a DNN that directly generates IMRT beam fluence maps from organ contours and dose distributions, bypassing traditional inverse planning.
• To assess the plan quality of DNN-generated fluence maps against clinically approved plans.

Main Methods:

• A DNN was trained using 240 prostate IMRT plans, utilizing organ contours and dose distributions as input.
• The trained DNN generated 45 synthetic plans (SPs) by creating beam fluence maps.
• SPs were compared to clinical plans (CPs) using plan quality metrics, including target homogeneity/conformity and organ-at-risk dose constraints (rectum, bladder, bowel).

Main Results:

• The DNN successfully generated fluence maps with minimal errors.
• The overall quality of the SPs was comparable to CPs.
• While target homogeneity was slightly reduced in SPs, conformity index, and critical organ dose constraints (V60Gy for rectum/bladder, V45Gy for bowel) showed no significant difference.

Conclusions:

• The proposed DNN method for direct fluence map generation is a viable next step in automating IMRT plan creation.
• This approach demonstrates potential to significantly improve treatment planning efficiency without compromising plan quality.
• The method facilitates maintaining high-quality radiotherapy plans through automated processes.