Keywords:
Not applicable, Observational, Artifacts, Computer Applications-General, CT, eHealth, Computer applications, Artificial Intelligence, Artificial Intelligence and Machine Learning
Authors:
M. Patwari1, R. Gutjahr2, R. Raupach3, A. Maier1; 1Erlangen/DE, 2Nürnberg, Deutschland/DE, 3Forchheim/DE
DOI:
10.26044/ecr2020/C-03463
Purpose
Due to the use of CT as a crucial medical tool for diagnostic and interventional purposes, the number of CT scans has been increasing. However, CT scans emit radiation which is harmful to the patient [1]. In accordance with the ALARA principle, we aim to reduce clinical CT dose to improve patient safety. However, reducing the radiation dose increases quantum noise in the reconstructed images [2]. Quantum noise in clinical CT is usually removed using nonlinear filters such as the bilateral filter [3], the median filter, or the guided filter [4]. Iterative reconstruction usually outperforms standard filtering [5,6], but is significantly more computationally intensive and time-consuming.
The use of convolutional neural networks (CNN) for low dose CT denoising has shown promising results in body scanning [7–9]. Current CNN denoising strategies optimise metrics such as the mean squared error, which causes smoothing of edges and the removal of low contrast features. Moreover, many denoising methods only operate on slices, and do not take information from neighboring slices into account.
In this study, we investigate the use of neighboring slices and perceptual metrics to train a denoising CNN. To observe the impact of neighboring slices, we use 3D networks for denoising [10, 11]. To investigate the impact of perceptual loss on noise removal and feature preservation, we introduce a CT image reconstruction quality network [12] into our optimization process.