Fiber Bundle Image Reconstruction Using Convolutional Neural Networks and Bundle Rotation in Endomicroscopy.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computational Science

Background:

• Fiber-bundle endomicroscopy suffers from the honeycomb effect, limiting image resolution and diagnostic accuracy.
• Existing methods struggle to fully resolve fine tissue structures, hindering clinical applications.

Purpose of the Study:

• To develop and validate a multi-frame super-resolution algorithm to mitigate the honeycomb effect in fiber-bundle endomicroscopy.
• To improve the quality and resolution of reconstructed tissue images for enhanced visualization.

Main Methods:

• A multi-frame super-resolution algorithm was developed, exploiting fiber-bundle rotation for feature extraction and tissue reconstruction.
• Simulated data with rotated fiber-bundle masks was used to train the machine learning model.
• The algorithm was trained and tested on prostate tissue images, with rigorous numerical analysis of super-resolved outputs.

Main Results:

• The algorithm successfully restored images with high quality, demonstrating a 1.97-fold improvement in mean structural similarity index (SSIM) over linear interpolation.
• The model exhibited robustness, performing well on test images without prior information.
• Image reconstruction was achieved rapidly (0.03s for 256x256 images), indicating potential for real-time application.

Conclusions:

• The novel combination of fiber bundle rotation and machine learning-based multi-frame image enhancement offers a significant improvement in image resolution for fiber-bundle endomicroscopy.
• This technique addresses a key limitation, paving the way for more precise and reliable in-vivo tissue imaging and diagnosis.