Sparse-View Photoacoustic Reconstruction Method for Diabetic Retinopathy Using Feature Fusion Network

Area of Science:

• Biomedical Optics
• Medical Imaging
• Ophthalmology

Background:

• Diabetic retinopathy is a leading cause of vision loss, stemming from diabetes mellitus complications.
• Photoacoustic imaging offers a noninvasive method for visualizing retinal vasculature, crucial for diagnosing diabetic retinopathy.
• Effective photoacoustic reconstruction is vital for high-quality image generation.

Purpose of the Study:

• To introduce a novel deep learning network, SAMF-Net, for enhanced photoacoustic image reconstruction.
• To evaluate SAMF-Net's performance using both raw photoacoustic signals and traditional reconstructions as inputs.
• To assess the network's capability in handling sparse detection views for improved image quality.

Main Methods:

• Developed a multi-input self-attention multiscale feature fusion network (SAMF-Net).
• The network integrates original photoacoustic signals and conventionally reconstructed images.
• Incorporated a self-attention mechanism for global feature extraction.

Main Results:

• SAMF-Net demonstrated superior photoacoustic reconstruction capabilities across various sparse detection scenarios.
• The proposed method effectively fused multi-scale features and global information.
• Qualitative and quantitative assessments confirmed the enhanced reconstruction quality.

Conclusions:

• SAMF-Net offers a significant advancement in photoacoustic image reconstruction techniques.
• The method shows promise for improving the diagnostic accuracy of diabetic retinopathy.
• This approach provides valuable insights for developing advanced medical imaging reconstruction algorithms.