Enhancing photoacoustic imaging for lung diagnostics and BCI communication: simulation of cavity structures artifact generation and evaluation of noise reduction techniques

Chengpeng Chai, Xi Yang, Xurong Gao, Junhui Shi, Xiaojun Wang, Hongfei Song, Yun-Hsuan Chen, Mohamad SawanFrontiers in Bioengineering and Biotechnology 2024 · DOI

TL;DR

This paper tackles a critical challenge in photoacoustic imaging (PAI) for lung diagnostics and brain-computer interfaces (BCI): artifacts caused by the lung's complex, cavity-filled structure. By developing a simulation model and evaluating noise reduction techniques like ADF and NLM, they demonstrate a path towards clearer PAI images, with experimental validation showing significant improvements in image quality.

What Problem Does This Paper Address?

Photoacoustic imaging (PAI) holds immense promise for non-invasively visualizing biological tissues, offering high resolution and deep penetration. It's particularly relevant for lung diagnostics (especially in light of pandemics like COVID-19) and for developing brain-computer interfaces (BCI). However, imaging structures like the lungs and brain with PAI is hampered by their inherent complexity, specifically the presence of air-filled cavities (e.g., alveoli, airways). These cavities disrupt the propagation of both light and sound waves, leading to artifacts that obscure crucial details like blood vessels. Traditional imaging methods like CT scans have limitations, such as ionizing radiation and a focus on structural changes rather than microvascular details. This paper aims to understand how these cavity-induced artifacts are generated and to evaluate methods for their removal to improve PAI's diagnostic and BCI capabilities.

Methods

The researchers employed a multi-faceted approach:

  • Simulation Model: They developed a simulation model using MATLAB to replicate the generation and propagation of photoacoustic signals within simplified lung-like cavity structures. This involved:
  • Optical Simulation: Using Monte Carlo methods to simulate light propagation and absorption, considering both uniform and non-uniform light source distributions.
  • Acoustic Simulation: Employing the k-space pseudo-spectral method to model sound wave propagation.
  • Image Reconstruction: Utilizing the Delay-and-Sum (DAS) algorithm.
  • Physical Verification: They constructed realistic phantoms mimicking lung cavity structures (using agarose, blood, water, and air-filled tubes) and imaged them using a PACT system.
  • Artifact Removal Algorithms: Two key noise reduction techniques were evaluated:
  • Anisotropic Diffusion Filter (ADF): An edge-preserving filter that adapts diffusion based on local image characteristics.
  • Non-Local Means (NLM) Filter: A filter that leverages similarity between image regions to reduce noise, particularly effective for repetitive patterns.
  • Evaluation Metrics: A combination of objective metrics (PSNR, SSIM, MSE, NAE) and subjective visual assessment was used to quantify the effectiveness of the artifact removal techniques under varying conditions, especially light source uniformity.

Datasets

The study did not explicitly mention using pre-existing datasets for training. Instead, they generated their own data through:

  • Simulations: Using a custom-built simulation environment in MATLAB, generating synthetic PAI data based on defined optical and acoustic properties of tissue components.
  • Physical Experiments: Acquiring PAI data from custom-made phantoms designed to mimic lung cavity structures. These phantoms served as a novel benchmark for evaluating artifact generation and removal in controlled, yet realistic, scenarios.

Key Insights and Ideas

  • Cavity Structures are the Primary Artifact Source: The presence of air-filled cavities, due to their distinct acoustic impedance and sound speed, is the dominant factor in generating artifacts in PAI of lung-like structures.
  • NLM Excels in Uniform Lighting, ADF in Non-Uniform: A crucial finding is the differential performance of the filters. NLM shows superior artifact removal under uniform light conditions, likely due to its strength in handling repetitive patterns. However, ADF proves more effective under non-uniform lighting, which is more representative of real-world scenarios.
  • Light Source Uniformity is Critical: The uniformity of the light source significantly impacts artifact removal performance. Highly focused or uneven illumination exacerbates artifacts, making filter selection more challenging.
  • Acoustic Impedance Mismatch is Key: The paper highlights that differences in acoustic impedance between tissues (e.g., blood vs. air) are the root cause of strong signal reflections, leading to characteristic artifacts like semicircular arcs.

Key Results

  • Significant Image Quality Improvement:
    • NLM filtering in simulations increased the Peak Signal-to-Noise Ratio (PSNR) by 11.33%.
  • ADF filtering in experimental validation increased PSNR by 18.1%.
  • Filter Performance Trade-offs:
    • In simulations with uniform light, NLM generally outperformed ADF in terms of PSNR and SSIM.
  • In real experiments with non-uniform light, ADF showed better performance, particularly in simple object arrangements, with a higher PSNR improvement (18.1% vs. 15.12%).
  • Artifact Characteristics: Semicircular artifacts were observed, particularly around air inclusions, with their size influenced by the distance between blood and air tubes. The use of a half-ring ultrasound array also contributed to the semicircular shape of these artifacts.

Limitations and Future Work

  • Simplified Model: The study used simplified models of cavity structures (cylindrical tubes) and assumed uniform tissue properties. Future work should incorporate more complex, realistic organ geometries and tissue heterogeneity (e.g., varying densities of lymph nodes, vascular structures).
  • Grüneisen Parameter: The Grüneisen parameter, crucial for photoacoustic wave generation, was set to a unit value and not explicitly modeled. Incorporating this and other thermodynamic parameters could refine the simulation's accuracy.
  • Deep Learning Potential: While diffusion and NLM filters were explored, the authors suggest that deep learning methods (like CNNs) could offer further improvements, though they face challenges related to large training dataset requirements.
  • Hybrid Filter Strategies: The paper proposes developing hybrid approaches that dynamically select between ADF and NLM, or even integrate them sequentially, to optimize performance across various lighting conditions.
  • Light Source Optimization: Further research is needed to understand the precise impact of light source parameters (like waist radius) on artifact removal and to develop strategies for optimizing light source distribution.

Summary updated

Abstract

Pandemics like COVID-19 have highlighted the potential of Photoacoustic imaging (PAI) for Brain-Computer Interface (BCI) communication and lung diagnostics. However, PAI struggles with the clear imaging of blood vessels in areas like the lungs and brain due to their cavity structures. This paper presents a simulation model to analyze the generation and propagation mechanism within phantom tissues of PAI artifacts, focusing on the evaluation of both Anisotropic diffusion filtering (ADF) and Non-local mean (NLM) filtering, which significantly reduce noise and eliminate artifacts and signify a pivotal point for selecting artifact-removal algorithms under varying conditions of light distribution. Experimental validation demonstrated the efficacy of our technique, elucidating the effect of light source uniformity on artifact-removal performance. The NLM filtering simulation and ADF experimental validation increased the peak signal-to-noise ratio by 11.33% and 18.1%, respectively. The proposed technique adds a promising dimension for BCI and is an accurate imaging solution for diagnosing lung diseases.
BibTeX
@inproceedings{chai2024enhancing,
  author = {Chengpeng Chai and Xi Yang and Xurong Gao and Junhui Shi and Xiaojun Wang and Hongfei Song and Yun-Hsuan Chen and Mohamad Sawan},
  title = {{Enhancing photoacoustic imaging for lung diagnostics and BCI communication: simulation of cavity structures artifact generation and evaluation of noise reduction techniques}},
  booktitle = {Frontiers in Bioengineering and Biotechnology},
  year = {2024},
  doi = {10.3389/fbioe.2024.1452865},
}
Figure