Denoising Low-Dose CT Images Using Noise2Noise and Evaluation of Hyperparameters

In computed tomography (CT), the quality of the image is directly related to the exposure of the patient during the scan. A reduction in exposure reduces the health risks for patients, however, an increase in noise compromises the image quality. This work examines the Noise2Noise framework, which requires only noisy image pairs for network training in order to minimize the noise in CT images. This study examines the effects of varying learning rates, batch sizes, epochs, and encoder-decoder network depths on a variety of loss functions and their parameters.