177 300

Cited 0 times in

MR Weighted Image Discrimination by Artificial Intelligence

Authors
 Giljae Lee  ;  Gyehwan Jin  ;  Hwunjae Lee  ;  Jaeeun Jung 
Citation
 ScholarGen Journal of Medical Imaging, Vol.3(11) : 33, 2019-12 
Journal Title
ScholarGen Journal of Medical Imaging
ISSN
 2635-4608 
Issue Date
2019-12
Keywords
Image processing ; Discrete Wavelet Transform ; MR pulse sequence ; T2 Weighted Image ; MR Molecular Imaging ; Magnetic nanoparticles
Abstract
In this study, we proposed a method of
learning neural networks by optimizing neural
network input parameters to discern MRI-weighted
images. To this end, we segmented the weighting
domain of MRI. In feature extraction, the original
image and segmented image were extracted by DWT,
respectively. A neural network was trained by inputting extracted
feature values. As a result of the experiment, it was
found that the R-value of the segmented image is
closer to 1 than the original image. The reason is that
the images obtained by segmenting the areas of the
weighted parts already have similarities. Also, it was
found that the similarity between T1 and T2
weighted images is high in the same combination,
and the similarity is relatively low in different
weighted images. The most important issue in
medical imaging is ensuring the confidence of
radiologists using artificial intelligence. To solve this
problem, it is of utmost importance that the algorithm
developer and radiological A neural network was trained by inputting extracted
feature values. As a result of the experiment, it was
found that the R-value of the segmented image is
closer to 1 than the original image. The reason is that
the images obtained by segmenting the areas of the
weighted parts already have similarities. Also, it was
found that the similarity between T1 and T2
weighted images is high in the same combination,
and the similarity is relatively low in different
weighted images. The most important issue in
medical imaging is ensuring the confidence of
radiologists using artificial intelligence. To solve this
problem, it is of utmost importance that the algorithm
developer and radiological A neural network was trained by inputting extracted
feature values. As a result of the experiment, it was
found that the R-value of the segmented image is
closer to 1 than the original image. The reason is that
the images obtained by segmenting the areas of the
weighted parts already have similarities. Also, it was
found that the similarity between T1 and T2
weighted images is high in the same combination,
and the similarity is relatively low in different
weighted images. The most important issue in
medical imaging is ensuring the confidence of
radiologists using artificial intelligence. To solve this
problem, it is of utmost importance that the algorithm
developer and radiological technologist work
together to provide a solution that is integrated with
the radiologist's workflow.
Files in This Item:
T201906607.pdf Download
DOI
10.31916/sjmi2019-01-02
Appears in Collections:
1. College of Medicine (의과대학) > Research Institute (부설연구소) > 1. Journal Papers
Yonsei Authors
Lee, Hwun Jae(이훈재)
URI
https://ir.ymlib.yonsei.ac.kr/handle/22282913/179830
사서에게 알리기
  feedback

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse

Links