Intelligent Noise Reduction: Seeing Through The Noise With Deep Learning Image Processing

By Josh Johnson, Canon Medical Components

In general radiographic images, noise is typically proportional to the square root of the signal. Conventional noise reduction has been done using rule-based processing to manually separate the signal and the noise in the images based on the image characteristics of CXDI detectors, leading to improvements in image noise and contrast-to-noise ratio. However, limits have been reached on the amount of noise that can be reduced using these methods, making further improvements difficult for low-dose areas of images.

Machine learning techniques are being utilized to overcome these challenges. This paper delves into the development, application, and performance of Canon's Intelligent Noise Reduction, a Convolutional Neural Network (CNN)-based image processing procedure designed to produce high-quality images with a reduced patient radiation dose. Through data and visual examples, explore the transformative impact of Intelligent Noise Reduction in overcoming the limitations of conventional approaches.