A Physics-Informed Neural Network for In Vivo Dosimetry Using Quantitative Radiacoustic Imaging.

Area of Science:

• Medical Physics
• Biomedical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate in vivo dosimetry is crucial for radiotherapy but currently lacks direct clinical measurement methods.
• Radiacoustic imaging (RAI) shows potential for dose monitoring by detecting acoustic waves from radiation, but has been limited to qualitative assessments.

Purpose of the Study:

• To develop and validate a quantitative RAI (qRAI) framework using a physics-informed neural network (PINN) for in vivo dose reconstruction.
• To enable real-time, quantitative dose mapping directly within patients during radiotherapy.

Main Methods:

• A physics-informed neural network (PINN) was developed, integrating acoustic wave physics and digital twins of radiation delivery and detection systems.
• The PINN reconstructs quantitative dose maps from limited-view acoustic data, calibrated against experimental and simulated dose references.
• Validation was performed across diverse scenarios, including water tanks, human torso phantoms, and FLASH electron therapy.

Main Results:

• The PINN-based qRAI framework successfully reconstructed quantitative dose maps in vivo.
• The method demonstrated superior robustness and generalizability compared to purely data-driven models, particularly in settings without experimental ground truth.
• Accurate dose mapping was achieved across various clinical and experimental setups.

Conclusions:

• PINN-based qRAI provides a powerful tool for real-time, adaptive, and quantitative in vivo dosimetry.
• This technology addresses the critical need for direct, accurate dose measurement during radiotherapy.
• The framework shows significant promise for improving radiotherapy safety and efficacy.