Acta Optica Sinica, Volume. 45, Issue 3, 0317001(2025)

Deep Learning-Based Denoising Algorithm for Photoacoustic Endoscopy Targeting Time-Domain Data

Minhao Li, Zhuojun Xie, Yang Tang, and Jiaying Xiao*
Author Affiliations
  • Department of Biomedical Engineering, School of Basic Medical Sciences, Central South University, Changsha 410083, Hunan , China
  • show less
    Figures & Tables(9)
    PAE/USE system diagram[16]. (a) Schematic diagram of the structure of the photoacoustic/ultrasonic endoscopic system; (b) experimental data collection of rabbit rectum in vivo
    Raw A-line data, results of noise reduction with 6-time averaging and 60-time averaging
    Simulated endoscopic system acquisition. (a) Acquisition method simulated by the simulated image; (b) original synthetic image; (c) image after adding noise to Fig. 3(a)
    Aline-UNet neural network model
    Aline-CNN neural network model
    Simulation data and denoising results of wavelet transform, singular value decomposition, UNet-2D, Wave-UNet, Aline-CNN, and Aline-UNet. (a) Original simulation image as the reference image; (b) simulation image after noise addition; (c) reconstructed image from denoised output using wavelet transform; (d) reconstructed image from denoised output using singular value decomposition; (e) reconstructed image from denoised output using UNet-2D; (f) reconstructed image from denoised output using Wave-UNet; (g) reconstructed image from denoised output using Aline-UNet; (h) reconstructed image from denoised output using Aline-CNN; (i) SSIM comparison result; (j) magnified image of the area indicated by arrows in S1‒S8; (k) average PSNR comparison result; (l) difference images between the areas indicated by arrows in R1‒R6
    Comparison of photoacoustic data denoising reconstruction results by Aline-CNN and Aline-UNet models. (a) Photoacoustic results of rectum collected once; (b) average results after six times; (c) average results after 60 times; (d) results of Aline-CNN model after denoising; (e) results of Aline-UNet model after denoising; (f) SSIM comparison results and corresponding PSNR comparison results within ROI 1 and ROI 2 regions images
    Comparison of ultrasonic data denoising reconstruction results by Aline-CNN and Aline-UNet models. (a) Ultrasonic results of rectal rectum collected once; (b) average results after 6 times; (c) average results after 60 times; (d) results of Aline-CNN model after denoising; (e) results of Aline-UNet model after denoising; (f) SSIM comparison results and corresponding PSNR comparison results within ROI 1 and ROI 2 regions images
    • Table 1. Memory occupied and the time consumed during training for four noise reduction models under the same training configuration

      View table

      Table 1. Memory occupied and the time consumed during training for four noise reduction models under the same training configuration

      ModelMemory size /MBTime consumed /s
      UNet-2D8450.21715
      Wave-UNet8323.82136
      Aline-UNet45.4349
      Aline-CNN18.0821
    Tools

    Get Citation

    Copy Citation Text

    Minhao Li, Zhuojun Xie, Yang Tang, Jiaying Xiao. Deep Learning-Based Denoising Algorithm for Photoacoustic Endoscopy Targeting Time-Domain Data[J]. Acta Optica Sinica, 2025, 45(3): 0317001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Medical optics and biotechnology

    Received: Sep. 21, 2024

    Accepted: Nov. 18, 2024

    Published Online: Feb. 19, 2025

    The Author Email: Xiao Jiaying (jiayingxiao@csu.edu.cn)

    DOI:10.3788/AOS241580

    CSTR:32393.14.AOS241580

    Topics