High-fidelity image reconstruction in multimode fiber imaging through the MITM-Unet framework

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Machine Learning

Background:

• Single multimode fibers offer advantages for ultra-thin imaging probes but suffer from noise due to uneven speckle illumination.
• High-precision reconstruction of complex grayscale images is challenging with traditional methods.

Purpose of the Study:

• To develop an optimized image reconstruction framework for single multimode fiber imaging.
• To address noise issues from uneven speckle illumination and improve grayscale image reconstruction.

Main Methods:

• Combined inverse transmission matrix approach with deep neural networks (Unet).
• Integrated multiple illumination conditions and increased target exposure to mitigate noise.
• Developed the Multi-speckle Illumination type Inverse Transmission Matrix-Unet (MITM-Unet) method.

Main Results:

• MITM-Unet significantly outperformed Single-speckle illumination type (SITM-Unet).
• MITM-Unet achieved a structural similarity index of 0.59 and Pearson correlation coefficient of 0.91.
• SITM-Unet achieved a structural similarity index of 0.38 and Pearson correlation coefficient of 0.77.

Conclusions:

• The MITM-Unet method effectively reconstructs high-quality grayscale images from single multimode fiber systems.
• This approach provides valuable insights for advancing compact wide-field endomicroscopic imaging.
• The study demonstrates the potential for improved imaging capabilities using multimode fibers.