A deep-learning model for one-shot transcranial ultrasound simulation and phase aberration correction

Area of Science:

• Medical Physics
• Biomedical Engineering
• Neuroscience

Background:

• Transcranial ultrasound is a non-invasive neuromodulation technique with diverse clinical applications.
• Skull heterogeneity poses challenges for accurate ultrasound focusing and energy delivery.
• Existing methods face a trade-off between computational speed and accuracy in phase aberration correction.

Purpose of the Study:

• Introduce TUSNet, a deep learning framework for efficient and accurate transcranial ultrasound computations.
• Enable rapid prediction of pressure fields and phase corrections for clinical workflows.
• Address the limitations of current focus prediction methods.

Main Methods:

• Developed TUSNet, an end-to-end neural network for predicting 2D transcranial ultrasound pressure fields and phase corrections.
• Trained TUSNet on a large dataset of synthetic and real skull CT segments.
• Benchmarked TUSNet against k-Wave for speed, accuracy, and pressure estimation.

Main Results:

• TUSNet computed pressure fields and phase corrections in 21 ms, over 1200x faster than k-Wave.
• Achieved 98.3% accuracy in peak pressure magnitude estimation.
• Demonstrated a mean focal positioning error of 0.18 mm.

Conclusions:

• Deep learning, via TUSNet, offers accurate and rapid estimation of transcranial ultrasound parameters.
• TUSNet shows promise for accelerating ultrasound treatment planning.
• Further experimental validation is needed for real-world clinical conditions.