Low dose CT denoising with residual convolutional networks and deep learned perceptual loss

Patwari M, Gutjahr R, Raupach R, Maier A (2020)

Publication Language: English

Publication Type: Conference contribution, Abstract of a poster

Publication year: 2020

Event location: Vienna, Austria AT

URI: https://epos.myesr.org/poster/esr/ecr2020/C-03463

DOI: 10.26044/ecr2020/C-03463

Abstract

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. The use of convolutional neural networks (CNN) for low dose CT denoising has shown promising results in body scanning. Current CNN denoising strategies optimise metrics like the mean squared error, which causes smoothing of edges. We investigate the use of perceptual metrics to train a denoising CNN.

Seven CT body scans were reconstructed using weighted filtered backprojection. Each volume was reconstructed with the standard clinical dose and with doses of 50%, 25%, 10% and 5% of the standard clinical dose to simulate reduced dose acquisitions. An eight layer residual CNN was used as our denoising network. The network was trained with mean squared error as well as with the use of a deep learned perceptual quality network. 2D and 3D variants of the denoising networks were trained to compare denoising performance. PSNR and SSIM are used to measure noise and feature preservation respectively.

The use of a perceptual loss function improved denoising and feature preservation in a 2D denoising network (SSIM = 0.7020, PSNR = 33.8423) against mean squared error denoising (SSIM = 0.6813, PSNR = 33.7544). Perceptual loss functions improved feature preservation in 3D denoising networks (SSIM = 0.6736, PSNR = 33.9524) against mean squared error denoising (SSIM = 0.6699, PSNR = 33.9605).

Using a perceptual loss function improves feature preservation in low dose CT denoising. Our deep learned network captures more information about CT image features and CT image noise compared to the mean squared error.

Authors with CRIS profile

Mayank Patwari Lehrstuhl für Informatik 5 (Mustererkennung) Andreas Maier Lehrstuhl für Informatik 5 (Mustererkennung)

Involved external institutions

Siemens AG, Healthcare Sector DE Germany (DE)

How to cite

APA:

Patwari, M., Gutjahr, R., Raupach, R., & Maier, A. (2020, July). Low dose CT denoising with residual convolutional networks and deep learned perceptual loss. Poster presentation at European Congress of Radiology, Vienna, Austria, AT.

MLA:

Patwari, Mayank, et al. "Low dose CT denoising with residual convolutional networks and deep learned perceptual loss." Presented at European Congress of Radiology, Vienna, Austria 2020.

BibTeX: Download