Machine Learning-Based Optimization of tFUS Transducer Positioning for Targeted Visual Cortex Neuromodulation.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Acoustics

Background:

• Transcranial focused ultrasound (tFUS) offers non-invasive neuromodulation for visual cortex stimulation.
• Accurate targeting is hindered by complex skull acoustics and computationally intensive modeling.

Purpose of the Study:

• To compare machine learning models for optimizing tFUS transducer positioning and focal length.
• To reduce computational load while maintaining high targeting accuracy for visual cortex stimulation.

Main Methods:

• Compared Neural Network, Random Forest, Linear Regression, and Support Vector Regression models.
• Utilized tFUS acoustic simulation data with CT-derived skull models from 32 subjects.
• Incorporated the SphereFit model to account for skull curvature in transducer positioning.

Main Results:

• Random Forest model showed superior performance with the lowest average focal point error (2.76 ± 0.57 mm).
• Machine learning approaches significantly reduced computational demands.
• High targeting accuracy was maintained for visual cortex stimulation.

Conclusions:

• Machine learning effectively optimizes tFUS transducer positioning for precise visual cortex targeting.
• This approach enhances non-invasive neuromodulation efficiency.
• Potential applications include vision restoration and neurological rehabilitation.