A Deep Learning-Based Approach to Characterize Skull Physical Properties: A Phantom Study.

Area of Science:

• Medical imaging
• Biomedical engineering
• Acoustics

Background:

• Transcranial ultrasound imaging is vital for brain studies but distorted by skull bone.
• Skull bone thickness and porosity significantly impact ultrasound aberration.
• Current methods for assessing skull properties rely on CT or MRI scans.

Purpose of the Study:

• To develop ultrasound-based methods for estimating skull thickness and porosity.
• To investigate the use of machine learning and deep learning for analyzing ultrasound signals for skull properties.
• To enable improved transcranial ultrasound imaging by addressing skull-induced signal distortion.

Main Methods:

• Utilized physical skull-mimicking phantoms with varying thickness and porosity.
• Extracted features from ultrasound signals transmitted through the phantoms.
• Applied machine learning (ML) and deep learning (DL) models to predict phantom characteristics.

Main Results:

• Both ML and DL models accurately predicted skull phantom thickness and porosity.
• The models demonstrated reasonable accuracy in characterizing diverse skull phantom properties.
• Feature extraction from ultrasound signals proved effective for inferring physical skull characteristics.

Conclusions:

• Ultrasound signal analysis with ML/DL can estimate skull physical properties.
• This approach offers a non-invasive alternative to CT/MRI for skull characterization.
• The findings support the development of advanced skull aberration correction techniques for transcranial ultrasound.