Low-dose CT with a deep convolutional neural network blocks model using mean squared error loss and structural similar loss

Yinjin Ma; Peng Feng; Peng He; Zourong Long; Biao Wei

doi:10.1117/12.2542662

20 December 2019 Low-dose CT with a deep convolutional neural network blocks model using mean squared error loss and structural similar loss

Yinjin Ma, Peng Feng, Peng He, Zourong Long, Biao Wei

Author Affiliations +

Proceedings Volume 11209, Eleventh International Conference on Information Optics and Photonics (CIOP 2019); 112090I (2019) https://doi.org/10.1117/12.2542662
Event: Eleventh International Conference on Information Optics and Photonics (CIOP 2019), 2019, Xi'an, China

Abstract

X-Ray computed tomography (CT) is one of the most popular imaging modality in the medical image analysis for clinical application. Meanwhile, the potential risk of X-Ray radiation dose to patients has attracted the public attention. Over the past decades, extensive efforts have been made for developing low-dose CT. However, X-Ray radiation dose reduction may result in increased noise and artifacts, which can significantly compromise the image quality and deteriorate the diagnostic performance. Hence, restoring CT image from low-dose CT and improving the diagnostic performance is a challenging for the vast researchers and developers. In this paper, a method based on deep learning techniques is proposed for low-dose CT noise reduction. Our method integrates convolutional neural network (CNN) blocks, residual learning, exponential linear units (ELUs) into a deep learning framework. Especially, loss of structural similarity index (SSIM) is combines to the final objective function to improve the image quality. Differs from general deep learning based denoising method, our deep CNN blocks architecture learning noise directly from original low-dose CT images, then restores denoised CT image by subtracting the obtained noise image from the original low-dose CT. After training patch by patch, the proposed method attains promising performance compared to state of the art traditional methods (non-local means and Block-matching 3D) and representative deep learning methods (primary three layers convolutional neural networks and residual encoder-decoder convolutional neural network) in visual effects and quantitative measurements. Extensive experiments was implemented for how choosing the coefficients of overall loss function and the number of CNN blocks. The experimental results demonstrate that our noise reduction method is effective for low-dose CT and potential clinic application.

Citation Download Citation

Yinjin Ma, Peng Feng, Peng He, Zourong Long, and Biao Wei "Low-dose CT with a deep convolutional neural network blocks model using mean squared error loss and structural similar loss", Proc. SPIE 11209, Eleventh International Conference on Information Optics and Photonics (CIOP 2019), 112090I (20 December 2019); https://doi.org/10.1117/12.2542662

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available