Transcranial adaptive aberration correction using deep learning for phased-array ultrasound therapy.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Neuroscience

Background:

• Transcranial focused ultrasound (FUS) offers therapeutic potential but is hindered by skull-induced aberrations.
• Accurate phase correction is computationally intensive, limiting real-time applications.
• Opening the blood-brain barrier (BBB) safely requires precise FUS focusing.

Purpose of the Study:

• To develop a rapid and accurate deep learning-based method for transcranial aberration correction.
• To enable adaptive focusing for safe and effective ultrasonic treatment via temporary BBB opening.
• To reduce the computational burden associated with traditional aberration correction techniques.

Main Methods:

• A combined approach involving pre-segmentation, k-Wave simulation, and a 3D U-net deep learning network.
• Utilizing skull sound speed samples and phase delay as model inputs within a nonlinear simulation environment.
• Training a phase prediction model for efficient and accurate aberration correction.

Main Results:

• The proposed method achieved focus volume and grating lobe levels closest to the time reversal method.
• Mean peak values were at least 77% of the time reversal method's results.
• Computational cost per sample was reduced to ≤0.05s, a 200-fold improvement over traditional methods.

Conclusions:

• The deep learning approach enables rapid, precise, and adaptive transcranial aberration correction.
• This method overcomes limitations of 2D simulations and traditional computational demands.
• It presents a novel, efficient option for transcranial focused ultrasound therapies, including temporary BBB opening for brain diseases.