Enhancement of digital radiography image quality using a convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Digital radiography systems are crucial for medical and security imaging but suffer from low spatial resolution and signal-to-noise ratio.
• Existing image enhancement techniques often fail to adequately address these limitations, impacting diagnostic accuracy and security effectiveness.

Purpose of the Study:

• To investigate the efficacy of a modified convolutional neural network (CNN) in enhancing the image quality of digital radiography systems.
• To assess the network's ability to generate high-resolution, low-noise images from low-quality inputs.

Main Methods:

• Development and application of a novel residual to residual (RTR) convolutional neural network (CNN).
• Evaluation of the RTR-CNN on a test dataset of 5 X-ray images.
• Comparison of the proposed method against traditional techniques like bicubic interpolation and 3D block-matching.

Main Results:

• The RTR-CNN significantly improved image quality, as evidenced by a peak signal-to-noise ratio (PSNR) increase of approximately 1.3 dB compared to traditional methods.
• The proposed method achieved highly efficient processing times, completing enhancement within one second.
• Experimental results showed a remarkable improvement in object structural details, increased resolution, and reduced noise.

Conclusions:

• The residual to residual (RTR) convolutional neural network effectively enhances image quality in digital radiography systems.
• This AI-driven approach offers a promising solution for improving spatial resolution and reducing noise in medical and security imaging.
• The RTR-CNN demonstrates practical utility for upgrading image acquisition from existing digital radiography systems.