Closing the gap in plan quality: Leveraging deep-learning dose prediction for adaptive radiotherapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Online adaptive radiotherapy (ART) faces challenges balancing treatment quality and efficiency.
• Current ART systems often re-optimize using original planning goals, limiting quality improvements.
• Modifying planning goals can be time-consuming, posing a barrier to high-quality adaptive plans.

Purpose of the Study:

• To enhance adaptive radiotherapy plan quality by predicting new planning goals using a deep-learning dose prediction model.
• To account for inter-fractional anatomical changes during online adaptive therapy.
• To leverage AI for improved treatment planning in head and neck cancer.

Main Methods:

• Clinical evaluation of fine-tuned patient-specific (FT-PS) models for dose prediction in 23 adaptive fractions for 15 head-and-neck cancer patients treated with Ethos ART.
• Comparison of original adapted plans (IOE-Auto Plan) with AI-guided re-optimized plans (AI-guided IOE Plan) using physician-approved contours.
• Retrospective re-optimization in Ethos based on AI-predicted planning goals for organs at risk (OARs).

Main Results:

• Significant dose savings observed in nine high-impact OARs, including the constrictor, parotid glands, submandibular glands, oral cavity, esophagus, mandible, and larynx.
• The AI-guided IOE Plan was preferred over the original adapted plan in 19 out of 23 cases during a blind observer study.
• Demonstrated feasibility of AI-driven goal prediction for adaptive radiotherapy planning.

Conclusions:

• AI dose prediction models can effectively predict optimal planning goals, improving adaptive radiotherapy plan quality.
• The proposed method is applicable to both online and offline ART, showing potential for enhanced head-and-neck cancer treatment outcomes.
• Preliminary results indicate the feasibility and benefit of integrating AI for adaptive treatment planning.