One-dimensional convolutional neural network for Jacobian in Diffuse Optical Tomography.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Medical Physics

Background:

• Diffuse Optical Tomography (DOT) inverse problems are computationally intensive.
• Calculating the Jacobian matrix is a significant bottleneck in DOT reconstruction.
• Existing methods, like non-linear least squares, require substantial computation time.

Purpose of the Study:

• To develop a computationally efficient method for solving the DOT inverse problem.
• To accelerate the reconstruction process in Diffuse Optical Tomography.
• To reduce the time required for Jacobian matrix computation in DOT.

Main Methods:

• A data-driven neural network approach was developed.
• Singular Value Decomposition (SVD) was used to compute the updated Jacobian.
• A Convolutional Neural Network (CNN) was trained to map boundary measurements to singular values.

Main Results:

• The proposed neural network method significantly reduced computation time compared to the Adjoint method.
• The reconstructed absorption coefficients were comparable to those obtained using the Adjoint method.
• The method demonstrated improved computational efficiency for DOT inverse problems.

Conclusions:

• The data-driven neural network approach offers a viable solution for accelerating DOT reconstruction.
• This method has the potential to improve the efficiency of Diffuse Optical Tomography in clinical settings.
• Further research may lead to faster and more accurate DOT imaging for medical applications.