OpenKBP: The open-access knowledge-based planning grand challenge and dataset.

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Knowledge-based planning (KBP) is crucial for personalized radiation therapy.
• Accurate dose prediction is essential for optimizing treatment efficacy and minimizing toxicity.
• Existing methods for dose prediction lack standardized comparison frameworks.

Purpose of the Study:

• To establish a fair and consistent platform for evaluating dose prediction methods in KBP research.
• To benchmark the performance of various KBP models using a large, open-source dataset.
• To foster innovation and accelerate progress in radiation therapy planning.

Main Methods:

• The OpenKBP 2020 AAPM Grand Challenge was organized to develop and compare dose prediction models.
• Participants used a dataset of 340 head-and-neck cancer patients for model training, validation, and testing.
• Models were evaluated using two metrics: a 3D dose distribution score and a dose-volume histogram (DVH) score.

Main Results:

• The challenge attracted 195 participants, with 44 teams submitting models during the validation phase.
• In the testing phase, 28 unique prediction methods were evaluated.
• Top models achieved significant improvements, with one model excelling in both dose and DVH scoring compared to runner-up models.

Conclusions:

• OpenKBP successfully created the first standardized platform for comparing KBP prediction methods.
• The challenge democratized KBP research by providing open-source data and metrics.
• Publicly available OpenKBP datasets will facilitate future benchmarking and advancements in radiation therapy planning.