Deep learning–based MR‐to‐CT synthesis: The influence of varying gradient echo–based MR images as input channels

Florkow, Mateusz C.; Zijlstra, Frank; Willemsen, Koen; Maspero, Matteo; Berg, C.A.T. Van den; Kerkmeijer, Linda G.W.; Castelein, René M.; Weinans, Harrie; Viergever, Max A.; Stralen, Marijn van; Seevinck, Peter R.

doi:10.1002/mrm.28008

Cited by 94 publications

(92 citation statements)

References 45 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One latest research by Florkow et al also proved that combining multiple MR images can improve the performance and repeatability. 30 Although multiple MR sequence images are preferred to generate sCT images, however, in most clinical situations, multiple MR sequence images may not be available considering the acquisition time such as on-line MR guided radiotherapy. In this case, only T1-weighted images can be collected to predict sCT images with a higher HU accuracy.…”

Section: Discussionmentioning

confidence: 99%

“…Another advantage for using multiple MR sequence images is to improve the robustness, since MR images are prone to suffer from various image artifacts such as swallowing artifacts, using multiple MR sequence images would reduce this risk. One latest research by Florkow et al also proved that combining multiple MR images can improve the performance and repeatability . Although multiple MR sequence images are preferred to generate sCT images, however, in most clinical situations, multiple MR sequence images may not be available considering the acquisition time such as on‐line MR guided radiotherapy.…”

Section: Discussionmentioning

confidence: 99%

“…With the development of deep learning in various fields, several studies used deep convolutional neural networks (CNNs) to predict sCT and achieved encouraging results. [22][23][24][25][26][27][28][29][30] Xiao Han et al 22 first used CNN with U-net architecture to establish the image-to-image mapping between T1-weighted MR and CT images and realized the generation of sCT for patients with brain tumor. Nie et al 23 developed a three-dimensional context-aware GAN model for brain and pelvic dataset with single-sequence MR images as input.…”

Section: Introductionmentioning

confidence: 99%

“…With the development of deep learning in various fields, several studies used deep convolutional neural networks (CNNs) to predict sCT and achieved encouraging results . Xiao Han et al first used CNN with U‐net architecture to establish the image‐to‐image mapping between T1‐weighted MR and CT images and realized the generation of sCT for patients with brain tumor.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Multi‐sequence MR image‐based synthetic CT generation using a generative adversarial network for head and neck MRI‐only radiotherapy

et al. 2020

Medical Physics

View full text Add to dashboard Cite

Purpose: The purpose of this study is to investigate the effect of different magnetic resonance (MR) sequences on the accuracy of deep learning-based synthetic computed tomography (sCT) generation in the complex head and neck region. Methods: Four sequences of MR images (T1, T2, T1C, and T1DixonC-water) were collected from 45 patients with nasopharyngeal carcinoma. Seven conditional generative adversarial network (cGAN) models were trained with different sequences (single channel) and different combinations (multi-channel) as inputs. To further verify the cGAN performance, we also used a U-net network as a comparison. Mean absolute error, structural similarity index, peak signal-to-noise ratio, dice similarity coefficient, and dose distribution were evaluated between the actual CTs and sCTs generated from different models. Results: The results show that the cGAN model with multi-channel (i.e., T1 + T2 + T1C + T1DixonC-water) as input to predict sCT achieves higher accuracy than any single MR sequence model. The T1-weighted MR model achieves better results than T2, T1C, and T1DixonC-water models. The comparison between cGAN and U-net shows that the sCTs predicted by cGAN retains additional image details are less blurred and more similar to the actual CT. Conclusions: Conditional generative adversarial network with multiple MR sequences as model input shows the best accuracy. The T1-weighted MR images provide sufficient image information and are suitable for sCT prediction in clinical scenarios with limited acquisition sequences or limited acquisition time.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Multi‐sequence MR image‐based synthetic CT generation using a generative adversarial network for head and neck MRI‐only radiotherapy

et al. 2020

Medical Physics

View full text Add to dashboard Cite

show abstract

“…10,11 In our opinion, it can currently be said that the most interesting applications of machine vision to medical imaging do not necessarily lie within diagnostics or other tasks that human experts can perform as well, but rather within tasks that humans cannot ordinarily perform. Examples for this are the extraction of radiomic features such as genomic alterations from magnetic resonance imaging (MRI), 12 conversion of musculoskeletal MRI to computed tomography, 13 or to reduce the amount of Gadolinium contrast agent for MRI scans. 14 Furthermore, machine vision can help to prevent wrong-level spine surgery.…”

mentioning

confidence: 99%

Data Mining in Spine Surgery: Leveraging Electronic Health Records for Machine Learning and Clinical Research

Staartjes

Stienen

2019

Neurospine

View full text Add to dashboard Cite

Advances in natural language processing (NLP) and unsupervised learning have recently enabled the long-expected synergy between true "big data" and analytics based on machine learning (ML). The ability to reliably generate structured data from unstructured electronic health records (EHRs) such as free text reports, document scans, or unlabelled medical imaging has on the one hand allowed development of algorithms based on until recently unseen amounts of data, spanning entire hospital or even national populations, and not just databases compiled by human experts. On the other hand, the capacity to primarily generate structured reports from unstructured raw EHRs has proven valuable for hospital analytics, epidemiological studies, and systematic reviews.In their narrative review, Schwartz et al. 1 report on the utilization of EHRs in spine surgery through ML techniques. The authors are to be commended for their detailed description of data types commonly found in EHRs, learning concepts to generate structured data (such as NLP and machine vision), applications of ML for prognosis and prediction, and finally the challenges inherent to using unstructured data from EHRs in medical practice and research. As the authors show, there is no question that ML is already starting to affect surgical practice relevantly in many aspects. Especially the advent of open source algorithms provided by today's tech giants have largely democratized the development of ML models, and as the authors summarize, this has led to an explosion of publications reporting such algorithms. Still, it is important to preserve the methodological quality of papers utilizing ML techniques, which is often not the case. For example, the authors touch on the issue of ensuring generalizability through robust training structures (i.e., some form of resampling) and external validation, before models are rolled out into clinical practice.We especially value that the authors discuss the problem of uninterpretable "black box" models. 2 Currently, many groups are applying complex ML algorithms to relatively small patient samples and for relatively simple tasks. While this might lead to slight benefits in model performance, these complex models (such as deep neural networks for nonimaging applications) are often typical "black box" models with a total loss of the ability to explain what factors lead the algorithm to make a certain decision. Explicability is -unfortunately -often traded in for a small and likely irrelevant increase in model accuracy. Especially in Neurospine 2019;16(4):654-656.

show abstract

A review on medical imaging synthesis using deep learning and its clinical applications

Wang

Lei

et al. 2020

J Applied Clin Med Phys

192

131

View full text Add to dashboard Cite

This paper reviewed the deep learning‐based studies for medical imaging synthesis and its clinical application. Specifically, we summarized the recent developments of deep learning‐based methods in inter‐ and intra‐modality image synthesis by listing and highlighting the proposed methods, study designs, and reported performances with related clinical applications on representative studies. The challenges among the reviewed studies were then summarized with discussion.

show abstract

Deep learning–based MR‐to‐CT synthesis: The influence of varying gradient echo–based MR images as input channels

Abstract: This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Cited by 94 publications

References 45 publications

Multi‐sequence MR image‐based synthetic CT generation using a generative adversarial network for head and neck MRI‐only radiotherapy

Multi‐sequence MR image‐based synthetic CT generation using a generative adversarial network for head and neck MRI‐only radiotherapy

Data Mining in Spine Surgery: Leveraging Electronic Health Records for Machine Learning and Clinical Research

A review on medical imaging synthesis using deep learning and its clinical applications

Contact Info

Product

Resources

About