Image restoration for ring-array photoacoustic tomography based on an attention mechanism driven conditional...

Area of Science:

• Biomedical Imaging
• Medical Image Processing
• Photoacoustic Tomography

Background:

• Ring-Array photoacoustic tomography (PAT) systems offer non-invasive biomedical imaging capabilities.
• PAT images frequently exhibit quality issues like blurring and streak artifacts due to imperfect imaging conditions.

Purpose of the Study:

• To develop an advanced image restoration method for enhancing the quality of Ring-Array PAT images.
• To address common image degradation problems such as blurring and artifacts.

Main Methods:

• A conditional generative adversarial network (CGAN) framework was employed for image restoration.
• A Residual Shifted Window Transformer Module (RSTM) with hybrid spatial and channel attention was integrated into the generator.
• A comprehensive loss function was designed to optimize pixel accuracy, detail preservation, and perceptual quality.
• A gamma correction module was incorporated to improve image contrast.

Main Results:

• The proposed method demonstrated significant improvements in image resolution.
• Restoration of overall image quality was achieved on both simulated and in vivo datasets.
• The integration of attention mechanisms and RSTM enhanced the generator's performance in artifact removal and detail enhancement.

Conclusions:

• The developed CGAN-based image restoration method effectively enhances Ring-Array PAT images.
• The approach successfully mitigates blurring and streak artifacts, improving diagnostic image quality.
• This technique holds promise for advancing non-invasive biomedical imaging applications.