Deep-Learning-Based Algorithm for the Removal of Electromagnetic Interference Noise in Photoacoustic Endoscopic Image...

Area of Science:

• Biomedical Optics
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Photoacoustic endoscopy (PAE) shows clinical promise but is hindered by electromagnetic interference (EMI) noise and limited signal-to-noise ratio (SNR).
• Unlike other imaging modalities, PAE's SNR is fundamentally limited by optical pulse energy safety constraints, and hardware limitations exacerbate EMI susceptibility.
• Existing methods struggle to mitigate EMI noise effectively in PAE systems, impeding technological advancement and clinical adoption.

Purpose of the Study:

• To investigate the feasibility of using deep learning for electromagnetic interference (EMI) noise removal in photoacoustic endoscopy (PAE) image processing.
• To compare the performance of different deep learning architectures against classical filtering methods for EMI noise reduction in PAE.
• To demonstrate the capability of an optimized deep learning model in generating high-fidelity PAE images for preclinical research.

Main Methods:

• Four fully convolutional neural network architectures (U-Net, Segnet, FCN-16s, FCN-8s) were evaluated for EMI noise removal in PAE data.
• A modified U-Net architecture was specifically developed and optimized for enhanced performance in denoising PAE images.
• The deep learning approach was benchmarked against traditional filter-based noise reduction techniques.

Main Results:

• The modified U-Net architecture demonstrated superior performance in removing EMI noise compared to other evaluated deep learning models and classical filters.
• The U-Net model successfully generated a denoised 3D vasculature map from PAE data.
• The denoised map clearly depicted fine, mesh-like capillary networks within the wall of a rat colorectum, showcasing the technique's resolution capabilities.

Conclusions:

• Deep learning, particularly a modified U-Net architecture, offers a powerful and effective strategy for mitigating EMI noise in photoacoustic endoscopy (PAE).
• This AI-driven approach significantly improves PAE image quality, enabling visualization of intricate vascular structures previously obscured by noise.
• The developed AI strategy holds broad applicability for photoacoustic tomography (PAT) systems, especially those with inherent low SNR and limited hardware-based EMI prevention, paving the way for improved diagnostic capabilities.