Effective Use Of Complex Values! Accuracy Improvement By Applying Complex Neural Networks In MRI　

3 main points
✔️ Potential for speeding up MRI scans
✔️ Utilizing intrinsic properties of data through complex neural networks
✔️ Comparison of accuracy with general real convolution in various conditions

Analysis of Deep Complex-Valued Convolutional Neural Networks for MRI Reconstruction
written by Elizabeth K. Cole, Joseph Y. Cheng, John M. Pauly, Shreyas S. Vasanawala
(Submitted 3 Apr 2020 (v1), last revised 12 May 2020 (this version, v4))
Comments: Published by arXiv
Subjects: Image and Video Processing (eess.IV); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Medical Physics (physics.med-ph)

Code

Introduction

Data obtained in the real world is not limited to the range of real values. The MRI data in this paper is one of them, and I think that there are much data in the range of complex values including the signal processing domain. Currently, general CNN's calculate data within the range of real values, and it is not possible to convolve the data while maintaining the relationship of complex values.

However, by keeping those relationships intact, convolution may lead to more accurate predictions. The MRIs discussed in this paper, but one of the challenges with MRIs is that they have a slow scan speed. Although there are various methods to complete the scan faster, the intrinsic nature of the data is such that the accuracy of MRI image reconstruction can be increased and the speed of the scan can be improved by taking advantage of the complex relationships.

In this paper, we use complex CNNs that take into account the relationship between complex values and investigate the effect of the complex CNNs on accuracy under various conditions by comparing them with general (real) CNNs.

What is a complex CNN?

Categories related to this article

• Neural Network