A fully automated machine-learning-based workflow for radiation treatment planning in prostate cancer

Area of Science:

• Radiation Oncology
• Medical Artificial Intelligence
• Prostate Cancer Treatment

Background:

• Radiotherapy planning for prostate cancer can be time-consuming and subject to inter-observer variability.
• Artificial intelligence (AI) and machine learning (ML) show potential to improve efficiency and consistency in radiotherapy planning.

Purpose of the Study:

• To assess the clinical feasibility of a fully automated ML-based "one-click" workflow for prostate cancer radiotherapy planning.
• To compare the automated workflow's performance against conventional planning methods.

Main Methods:

• A fully automated ML workflow integrating segmentation and treatment planning was evaluated.
• The workflow was tested on five low-risk prostate cancer patients undergoing external beam radiotherapy.
• Results were compared with conventional plans based on multiple expert contours, analyzing qualitative and quantitative metrics.

Main Results:

• Deep-learning segmentation showed good agreement with expert contours (Dice Similarity, Hausdorff index).
• Automated plans achieved clinically acceptable dose coverage within the inter-observer variability range of manual plans.
• Automated segmentation resulted in smaller CTV volumes, leading to significantly lower dose coverage for CTV and PTV compared to manual plans.

Conclusions:

• The automated ML workflow is clinically feasible for prostate cancer radiotherapy, offering comparable results to conventional plans.
• This automation represents a step towards more efficient and standardized cancer treatment.
• Further improvements in segmentation models and prospective testing are necessary due to observed smaller target volumes with auto-segmentation.