Automatic planning for head and neck seed implant brachytherapy based on deep convolutional neural network dose...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence in Medicine

Background:

• Seed implant brachytherapy (SIBT) is a key treatment for head and neck (H&N) cancers.
• Current SIBT planning faces challenges with manual needle path setting and inaccurate dose calculations.

Purpose of the Study:

• To develop a precise deep convolutional neural network dose engine (DCNN-DE) for H&N SIBT.
• To create an automated SIBT planning method for improved efficiency and accuracy.

Main Methods:

• A 3D-unet based DCNN-DE was developed, incorporating a novel inter-seed shadow map (ISSM) for accurate dose prediction.
• An automated planning strategy generated needle paths avoiding critical structures, optimized using a heuristic method.
• The DCNN-DE was validated against Monte Carlo simulations (MCS).

Main Results:

• DCNN-DE reduced TG-43 calculation errors for CTV V100 and D90 by 93% and 92% respectively, achieving accuracy close to MCS.
• Automated planning generated clinically acceptable plans in an average of 2.5 minutes.
• Dose distributions in generated plans were comparable to manually created clinical plans.

Conclusions:

• The developed DCNN-DE and automated planning method offer high accuracy and efficiency for H&N SIBT.
• This approach shows significant potential for clinical adoption, improving SIBT treatment planning.
• The method provides rapid generation of clinically acceptable SIBT plans with precise dose evaluation.