Using multi-centre data to train and validate a knowledge-based model for planning radiotherapy of the head and neck

Area of Science:

• Radiation Oncology
• Medical Physics
• Computational Biology

Background:

• Knowledge-based radiotherapy planning (KB RPT) models enhance treatment precision.
• Larger training datasets improve the robustness of KB RPT models.
• Combining models from multiple institutions can broaden training knowledge.

Purpose of the Study:

• To develop a method for merging knowledge-based radiotherapy planning models from multiple centers into a single 'super-model'.
• To increase the breadth and robustness of training knowledge for improved treatment planning.

Main Methods:

• A head and neck super-model was created by merging 207 patient datasets from three centers.
• The super-model was validated on 30 independent datasets and tested on 40 unseen patients from four centers.
• Generated plans were evaluated using established criteria for plan quality and efficiency.

Main Results:

• The super-model generated plans meeting dose objectives in an average of 10 minutes.
• Significant improvements in healthy tissue sparing were observed: parotid (4.7 Gy), spinal cord (3.3 Gy), and brainstem (2.9 Gy).
• Target coverage met constraints, with a marginal reduction compared to clinical plans.

Conclusions:

• Merging patient libraries from multiple centers successfully created a super-model for head and neck radiotherapy planning.
• The super-model demonstrated improved healthy tissue sparing and met evaluation criteria.
• This approach has the potential to enhance the quality, efficiency, and consistency of radiotherapy planning across institutions.