Evaluation of a prostate SBRT planning workflow using auto-segmentation and knowledge-based planning

Area of Science:

• Radiation Oncology
• Medical Physics
• Artificial Intelligence in Healthcare

Background:

• Prostate stereotactic body radiotherapy (SBRT) demands precise target delineation and organ at risk (OAR) contouring for optimal treatment planning.
• Integrating artificial intelligence (AI) contouring into SBRT workflows offers potential for increased efficiency and accuracy.
• Knowledge-based planning (KBP) is a sophisticated method for automating radiotherapy plan generation.

Purpose of the Study:

• To evaluate the comparability of AI-generated contours versus physician-delineated contours in achieving dosimetric goals for prostate SBRT.
• To assess the accuracy of commercial AI contouring software for prostate, rectum, and bladder delineation.
• To determine the dosimetric impact of AI-generated OAR contours within a KBP framework.

Main Methods:

• Retrospective analysis of 20 prostate cancer patients undergoing SBRT.
• Application of commercial AI contouring software to CT scans for prostate, rectum, and bladder.
• Comparison of AI contours with physician-delineated contours using Dice Similarity Coefficient (DSC), surface DSC (sDSC), and Added Path Length (APL).
• Generation of Volumetric Modulated Arc Therapy (VMAT) plans using an in-house prostate SBRT KBP model with different contouring inputs (clinical, AI, post-processed AI).

Main Results:

• AI prostate contours were deemed clinically unacceptable.
• AI rectum and bladder contours showed overlap with clinical prostate contours in a significant number of cases.
• While AI OARs initially caused minor dosimetric deviations, post-processing and recalculation on clinical contours led to acceptable plan quality, comparable to reference plans.
• Plans utilizing post-processed AI OARs demonstrated improved rectum and bladder sparing compared to initial AI OARs.

Conclusions:

• A fully automated AI contouring and planning workflow for prostate SBRT is not yet feasible.
• Physician-delineated target volumes remain essential for accurate prostate SBRT planning.
• Combining physician-delineated target volumes with post-processed AI contours for OARs shows encouraging results for achieving dosimetric goals in prostate SBRT.