Technical Note: Dose prediction for head and neck radiotherapy using a three-dimensional dense dilated U-net...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Radiation therapy treatment planning is a manual, time-consuming process with variable plan quality.
• Artificial intelligence (AI) offers potential for improving radiotherapy plan quality and efficiency.

Purpose of the Study:

• To describe participation in the American Association of Physicists in Medicine Open Knowledge-Based Planning Challenge (OpenKBP).
• To accurately predict radiation therapy dose distributions using AI.

Main Methods:

• Development of a 3D densely connected U-Net with dilated convolutions for dose prediction.
• Utilized contoured CT images of head and neck patients as input.
• Employed a custom-weighted mean squared error loss function and an ensemble of networks.

Main Results:

• Achieved second place in the OpenKBP challenge dose stream.
• Average mean absolute difference between predicted and clinical dose distributions was 2.56 Gy.
• Predicted DVH metrics for targets and organs at risk were within 3% and 2 Gy of clinical plans, respectively.

Conclusions:

• The 3D dense dilated U-Net accurately predicts 3D radiotherapy dose distributions.
• This AI model can be integrated into automated radiation therapy planning pipelines.