Virtual-simulation boosted neural network dose calculation engine for intensity-modulated radiation therapy.

Area of Science:

• Medical Physics
• Artificial Intelligence in Healthcare
• Radiotherapy Dosimetry

Background:

• Monte Carlo (MC) dose calculation is the benchmark for accuracy but is computationally intensive.
• Rapid and precise dose calculation is crucial for advanced radiation therapy techniques like IMRT.
• Existing methods often face a trade-off between speed and accuracy.

Purpose of the Study:

• To develop a rapid and accurate dose calculation engine using a neural network (NN) trained on a virtual simulation database.
• To achieve dose distribution accuracy comparable to MC methods while drastically reducing computation time.
• To establish an automated workflow for NN model training in fixed-beam intensity-modulated radiation therapy (IMRT).

Main Methods:

• Established a virtual simulation database using automated optimization techniques to generate individual beam dose distributions.
• Constructed and trained a 3D Dense-U-Net neural network architecture using the generated dataset.
• Validated the NN model's accuracy on IMRT plans for nasopharyngeal, cervical, and lung cancers.

Main Results:

• The NN model achieved significantly improved gamma passing rates (1 mm/1% and 2 mm/2%) for clinical beam doses, reaching 77.5% and 95.6%, respectively.
• Mean computation time for dose calculation was drastically reduced to 0.017 ± 0.002 seconds.
• The automated workflow successfully generated a large training dataset from a smaller clinical dataset, enabling high model performance.

Conclusions:

• A novel neural network-based dose calculation model for fixed-beam IMRT has been successfully developed.
• The model demonstrates high accuracy and exceptional speed, offering a viable alternative to traditional MC methods.
• The automated training workflow facilitates the creation of accurate and efficient AI-driven radiotherapy tools.