Direct integration of deep learning-based GTV auto-segmentation into a clinical radiotherapy planning system

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Manual Gross Tumor Volume (GTV) delineation in radiotherapy is labor-intensive and suffers from inter-observer variability.
• Deep Learning (DL) based auto-segmentation promises efficiency but faces challenges in clinical integration.
• Bridging the gap between DL auto-segmentation and clinical radiotherapy workflow is crucial.

Purpose of the Study:

• To integrate an in-house DL GTV auto-segmentation model into a radiotherapy Treatment Planning System (TPS).
• To demonstrate the seamless deployment and operational feasibility of DL-based GTV segmentation within a clinical TPS.
• To assess the time efficiency and workflow impact of integrated DL auto-segmentation.

Main Methods:

• Development and validation of an in-house DL model for GTV auto-segmentation.
• Integration of the DL model into a commercial radiotherapy Treatment Planning System (TPS).
• Demonstration using a head-and-neck GTV segmentation use case.

Main Results:

• Successful and seamless integration of the DL GTV auto-segmentation model into the TPS.
• DL-generated GTV contours were produced in under three minutes.
• The integration did not disrupt the established radiotherapy workflow.

Conclusions:

• DL-based GTV auto-segmentation can be effectively integrated into radiotherapy TPS.
• This integration offers a feasible solution to reduce manual delineation time and inter-observer variability.
• The developed system demonstrates potential for enhancing radiotherapy treatment planning efficiency and consistency.