Deep learning improves contrast in low-fluence photoacoustic imaging

Area of Science:

• Biomedical Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Low fluence illumination sources offer practical advantages for photoacoustic imaging (PAI) clinical use, including portability and safety.
• However, low fluence negatively impacts PAI image quality, limiting diagnostic accuracy.
• Developing methods to overcome low fluence limitations is crucial for PAI's clinical adoption.

Purpose of the Study:

• To develop and validate a novel denoising method for improving photoacoustic image quality obtained with low fluence illumination.
• To map low fluence PAI images to high fluence equivalents using a deep learning approach.
• To assess the effectiveness of the proposed method in enhancing image quality metrics and contrast.

Main Methods:

• A multi-level wavelet-convolutional neural network (CNN) was designed for image denoising.
• The CNN was trained to transform low fluence PAI images into high fluence excitation maps.
• Quantitative metrics (PSNR, SSIM, CNR) and qualitative assessments were used for evaluation, including an in vivo study.

Main Results:

• The proposed wavelet-CNN method significantly reduced background noise while preserving target structures.
• Substantial improvements were observed in Peak Signal-to-Noise Ratio (PSNR) by up to 2.20-fold, Structural Similarity Index Measure (SSIM) by up to 2.25-fold, and Contrast-to-Noise Ratio (CNR) by up to 4.3-fold.
• In vivo imaging demonstrated an enhanced contrast by up to 1.76-fold.

Conclusions:

• The developed denoising tool effectively enhances photoacoustic image quality from low fluence sources.
• This method holds significant potential for improving the clinical utility of portable and affordable PAI systems.
• The proposed technique can increase the value of low fluence illumination sources in photoacoustic imaging applications.