Accurate and Fast Deep Learning Dose Prediction for a Preclinical Microbeam Radiation Therapy Study Using...

Area of Science:

• Medical Physics
• Radiotherapy
• Computational Biology

Background:

• Microbeam radiation therapy (MRT) shows promise for treating aggressive tumors like gliosarcomas, but precise dose calculation is challenging.
• Current Monte Carlo (MC) simulations for MRT are computationally intensive, hindering treatment optimization and adaptive radiotherapy.
• Existing fast dose estimation methods are not available for emerging radiotherapy techniques like MRT.

Purpose of the Study:

• To develop and validate a fast and accurate machine learning (ML) model for predicting dose distributions in preclinical microbeam radiation therapy (MRT).
• To enable efficient dose calculation for novel cancer radiotherapy applications, overcoming limitations of traditional MC simulations.

Main Methods:

• Generated digital phantoms from CT data for preclinical rodent models.
• Used Geant4 to simulate MRT dose deposition with high-noise (15%) and low-noise (2%) MC simulations.
• Trained an ML model on high-noise MC data to predict peak and valley doses within phantoms.
• Validated the ML model's predictions against low-noise MC simulations.

Main Results:

• The ML model achieved high accuracy, with predictions within 3% of low-noise MC simulations for a significant majority of voxels.
• Valley dose predictions agreed with MC simulations for at least 77.6% of all voxels (95.9% of tumor voxels).
• Peak dose predictions agreed with MC simulations for at least 93.9% of all voxels (100.0% of tumor voxels).

Conclusions:

• Successfully demonstrated the first application of a fast ML dose prediction model for preclinical MRT.
• Utilizing high-noise MC simulations for training significantly accelerates data generation for ML models.
• The developed ML approach is transferable to other treatment modalities, advancing novel radiation cancer therapies.