Key points
-
This article aims to provide a comprehensive review on the applications of generative adversarial networks (GANs) in brain MRI.
-
The specific focus of this education review is on brain MRI.
-
It covers a large number of studies on GANs in brain MRI and the most recently published studies on brain MRI.
Introduction
Previous review | Year | Scope and coverage | Comparative contribution of our review |
---|---|---|---|
An overview of deep learning in medical imaging focusing on MRI [1] | 2019 | (1) It did not focus on GANs but rather covered many different deep learning methods (2) It did not focus on just brain MRI but rather focused on different MRI (3) It did not cover many recent studies as there has been an exponential rise in GANs-based methods for brain MRI during the last 2 years | (1) Our review is focused on GANs (2) Our review is focused on brain MRI (3) Our review covers many recent studies, published in 2020 and 2021 |
Review of deep learning approaches for the segmentation of multiple sclerosis lesions on brain MRI [2] | 2020 | (1) It did not focus on GANs but rather covered a broad range of deep learning methods (2) It did not cover applications for brain MRI such as synthesis of brain MRI data, translation of brain MRI data, and deep learning for noise removal from brain MRI, etc. | (1) Our review is focused on GANs (2) Our review covers all the possible applications for brain MRI |
Generative adversarial network in medical imaging: A review [3] | 2019 | (1) It did not focus on brain MRI but rather covered all modalities of medical imaging (2) It did not cover many recent studies published in 2020 and 2021, as there has been an exponential rise in studies for brain MRI during the last 2 years | (1) Our review is focused on brain MRI (2) Our review covers many recent studies, published in 2020 and 2021 |
Methods
Search strategy
Search sources
Search terms
Search eligibility criteria
Study selection
Data extraction
Data synthesis
Results
Search and study selection results
Demographics of the included studies
Number of studies | |
---|---|
Year | |
Year | |
2022 | 1 |
2021 | 44 |
2020 | 60 |
2019 | 28 |
2018 | 5 |
2017 | 1 |
Countries | |
Country | |
China | 53 |
USA | 22 |
Japan | 11 |
Germany | 7 |
India | 7 |
South Korea | 6 |
France | 4 |
Sweden | 3 |
Israel | 3 |
Canada | 3 |
Australia | 2 |
UK | 2 |
Singapore | 2 |
The Netherlands | 2 |
Italy | 2 |
United Arab Emirates | 1 |
Turkey | 1 |
Switzerland | 1 |
Spain | 1 |
Russia | 1 |
Malaysia | 1 |
Jordan | 1 |
Ireland | 1 |
Iran | 1 |
Malaysia | 1 |
Type of publication | |
Venue | |
Conference | 52 |
Journal | 87 |
Applications of GANs in brain MRI
Applications of studies | No. of studies | Reference of the study |
---|---|---|
Applications of studies | ||
Synthesis | 43 | |
Segmentation | 32 | |
Diagnosis | 22 | |
Reconstruction | 13 | |
Super-resolution | 10 | |
Prediction | 7 | |
Noise removal | 5 | |
Prognosis | 4 | |
Image registration | 2 | |
3D synthesis | 1 | [38] |
Purpose of using GANs | ||
Synthesis | 45 | |
Segmentation | 26 | |
Diagnosis | 16 | |
Reconstruction | 15 | |
Translation | 12 | |
Super-resolution | 7 | |
Noise removal | 5 | |
Prediction | 5 | |
Prognosis | 4 | |
Features extraction | 1 | [39] |
Translation | 1 | |
Anomaly detection | 1 | [94] |
Image registration | 1 | [144] |
Types of GANs methods
Types of datasets
Dataset name | URL | No. of studies | IDs of studies |
---|---|---|---|
Alzheimer’s Disease Neuroimaging Initiative (ADNI) | 16 | ||
BRATS2018 | 8 | ||
IXI dataset | 7 | ||
BRATS2016 | 4 | ||
Connectome | 3 | ||
BrainWeb | 3 | ||
Decathlon | 3 | ||
Figshare | 3 | ||
3 | |||
BRATS 2013 | 2 | ||
BraTS 2015 | 2 | ||
BraTS 2017 | 2 | ||
HCP | 2 | ||
Cancer Imaging | 2 | ||
PPMI | 2 | ||
2 | |||
Brats 2014 | 1 | [142] | |
Brats 2019 | 1 | [76] | |
ISLES | 1 | [8] | |
NAMIC dataset | 1 | [9] | |
MIT | 1 | [23] | |
MRIdata | 1 | [36] | |
Harvard | 1 | [82] | |
VIM | 1 | [40] | |
BIT China | 1 | [60] | |
CIND | 1 | [80] | |
IBSR | 1 | [113] | |
Hisub | 1 | [25] | |
ATAG | 1 | [115] | |
Cabal | 1 | [135] | |
John Hopkins University | 1 | [125] | |
CSIRO | 1 | [132] | |
NIFD | 1 | [6] | |
GDC | 1 | [98] | |
UK Data Service | 1 | [102] | |
NFB | 1 | [102] | |
ISEG2017 | 1 | [113] | |
OpenNeuro | 1 | [127] | |
ATLAS dataset | http://fcon_1000.projects.nitrc.org/indi/retro/atlas.html | 1 | [54] |
OpenNeuro2 | 1 | [122] |
Evaluation procedure
Evaluation mechanism | Number of studies | IDs of studies |
---|---|---|
Train, validate, test split | 17 | |
Training, test split | 38 | |
Twofold cross-validation | 3 | |
Threefold cross-validation | 2 | |
Fourfold cross-validation | 2 | |
Fivefold cross-validation method | 12 | |
Sevenfold cross-validation | 2 | |
Tenfold cross-validation | 6 | |
External | 7 |
Evaluation metric | Number of studies | IDs of studies |
---|---|---|
SSIM | 53 | |
PSNR | 49 | |
DSC | 31 | |
Accuracy | 22 | |
MAE | 16 | |
MSE | 16 | |
Sensitivity | 11 | |
Precision | 9 | |
Recall | 9 | |
F1 score | 8 | |
FID | 8 | |
Specificity | 8 |
Focal diseases in the studies
Focal disease | Number of studies (n) | IDs of studies |
---|---|---|
Brain tumor | 44 | |
Neurodegenerative disorders | 20 | |
None | 75 |