Enhanced model iteration algorithm with graph neural network for diffuse optical tomography.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Science

Background:

• Diffuse Optical Tomography (DOT) uses near-infrared light to image biological tissues.
• DOT reconstruction is challenging due to photon scattering and limited data, requiring improved accuracy and computational efficiency.
• The Levenberg-Marquardt (LM) algorithm is common but computationally intensive and can lack precision.

Purpose of the Study:

• To develop a novel neural network-based iterative algorithm for enhanced DOT reconstruction.
• To evaluate the performance of Graph Convolutional Neural Network (GCN) and Graph Attention Neural Network (GAT) integrated with LM.
• To assess the generalization capability of the proposed methods for accurate optical parameter retrieval.

Main Methods:

• A Graph Neural Network combined with Levenberg-Marquardt (GNNLM) algorithm was proposed.
• Graph data structures represented finite element meshes for DOT reconstruction.
• Two GNN variants, GCNLM and GATLM, were experimentally evaluated using simulations and phantom studies.

Main Results:

• GCNLM performed best within the training data distribution in simulations.
• GATLM demonstrated superior performance outside the training data distribution and in real phantom experiments.
• GATLM showed excellent generalization from simulation to real-world data without transfer learning.

Conclusions:

• GATLM offers a robust and accurate method for Diffuse Optical Tomography reconstruction.
• The GATLM algorithm generalizes well, improving absorption coefficient distribution accuracy in clinical settings.
• This approach holds promise for more precise non-invasive tissue optical property measurement.