Regularized Dual Averaging Image Reconstruction for Full-Wave Ultrasound Computed Tomography

Area of Science:

• Medical Imaging
• Computational Physics
• Biomedical Engineering

Background:

• Ultrasound Computed Tomography (USCT) shows potential for breast cancer screening.
• Waveform inversion methods yield high-resolution USCT images but are computationally intensive.
• Current methods combine stochastic and deterministic terms, limiting optimization.

Purpose of the Study:

• To introduce a structured optimization method for USCT image reconstruction.
• To improve regularization effectiveness and maintain reduced reconstruction times.
• To enable incorporation of nonsmooth regularization penalties.

Main Methods:

• Combined dual averaging with source encoding techniques for USCT image reconstruction.
• Decomposed iterations into gradient descent and proximal update steps.
• Utilized a time-domain method to solve the wave equation, avoiding explicit differentiation of regularization terms.

Main Results:

• The dual averaging method improved regularization effectiveness.
• Reconstruction times were maintained at reduced levels due to source encoding.
• Generated USCT images exhibited less noise and comparable resolution to Stochastic Gradient Descent (SGD) methods.

Conclusions:

• The dual averaging method offers an effective approach for USCT image reconstruction.
• This method enhances image quality by reducing noise.
• It provides a viable alternative to SGD for computationally demanding USCT applications.