Super-resolution technology to simultaneously improve optical & digital resolution of optical coherence tomography...

Area of Science:

• Medical Imaging
• Optical Coherence Tomography (OCT)
• Deep Learning

Background:

• Optical Coherence Tomography (OCT) is vital for cardiac imaging, particularly for plaque assessment before and after stenting.
• Improving OCT image resolution is crucial, but existing research primarily focuses on digital resolution, neglecting optical resolution improvements.
• High-resolution OCT imaging presents challenges in optical design, data storage, and transmission.

Purpose of the Study:

• To develop a deep learning method for generating high-resolution (HR) OCT images from low optical and low digital resolution (L²R) images.
• To enhance both optical and digital resolutions simultaneously in OCT cardiac imaging.
• To evaluate the effectiveness of the proposed method in reconstructing accurate and high-quality OCT images.

Main Methods:

• Modification of the Super-Resolution Generative Adversarial Network (SR-GAN) architecture for OCT image reconstruction.
• Application of the modified SR-GAN to generate HR OCT images from L²R inputs.
• Comparative analysis of the proposed method against established denoising techniques like BM3D and DnCNN.

Main Results:

• Reconstructed OCT images from highly compressed data demonstrated high structural similarity and accuracy compared to HR images.
• The deep learning method successfully improved both optical and digital resolutions of OCT images.
• The proposed method exhibited superior denoising performance compared to BM3D and DnCNN.

Conclusions:

• The developed deep learning approach effectively reconstructs high-resolution OCT cardiac images from low-resolution inputs.
• This method addresses the limitations of improving both optical and digital resolutions in OCT imaging.
• The technique offers a promising solution for enhanced cardiac OCT analysis, improving diagnostic accuracy and image quality.