Lightweight sparse optoacoustic image reconstruction via an attention-driven multi-scale wavelet network.

Area of Science:

• Biomedical Imaging
• Medical Technology
• Computational Imaging

Background:

• Photoacoustic tomography (PAT) offers high-contrast, high-resolution, and rapid biomedical imaging.
• Image quality in PAT is sensitive to sampling density; sparse sampling reduces costs but introduces artifacts.
• Existing deep learning models for sparse PAT imaging are computationally intensive, limiting their use in resource-constrained settings.

Purpose of the Study:

• To develop a computationally efficient deep learning model for high-quality sparse PAT image reconstruction.
• To address the limitations of existing methods in terms of computational complexity and resource requirements.

Main Methods:

• Introduction of AD-WaveNet, a lightweight neural network.
• Integration of the Discrete 2D Wavelet Transform (DWT) with adaptive attention mechanisms.
• Attention mechanisms designed to leverage DWT's multi-scale decomposition for feature emphasis.

Main Results:

• AD-WaveNet significantly enhances sparse PAT image reconstruction quality.
• The proposed model drastically reduces computational complexity and parameter count (nearly two orders of magnitude) compared to state-of-the-art methods.
• Optimal reconstruction quality is maintained despite reduced computational load.

Conclusions:

• AD-WaveNet offers a highly efficient solution for sparse PAT imaging.
• The network's lightweight design and effective attention mechanisms make it suitable for resource-constrained environments.
• Demonstrates significant potential for practical implementation in PAT applications.