Back

Unsupervised Harmonization Of Brain MRI Using 3D Cycle Gans And Its Effect On Brain Age Prediction

Komandur, D.; Gupta, U.; Chattopadhyay, T.; Dhinagar, N.; Thomopoulos, S. I.; Chen, J.-C.; Beavers, D.; Steeg, G. v.; Thompson, P. M.

2022-11-15 neuroscience
10.1101/2022.11.15.516349 bioRxiv
Show abstract

Deep learning methods trained on brain MRI data from one scanner or imaging protocol can fail catastrophically when tested on data from other sites or protocols - a problem known as domain shift. To address this, here we propose a domain adaptation method that trains a 3D CycleGAN (cycle-consistent generative adversarial network) to harmonize brain MRI data from 5 diverse sources (ADNI, WHIMS, OASIS, AIBL, and UK Biobank; total N=4,941 MRIs, age range: 46-96 years). The approach uses 2 generators and 2 discriminators to generate an image harmonized to a specific target dataset given an image from the source domain distribution and vice versa. We train the CycleGAN to jointly optimize an adversarial loss and cyclic consistency. We use a patch-based discriminator and impose identity loss to further regularize model training. To test the benefit of the harmonization, we show that brain age estimation - a common benchmarking task - is more accurate in GAN-harmonized versus raw data. t-SNE maps show the improved distributional overlap of the harmonized data in the latent space.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

1
Medical Image Analysis
○ 35 papers in training set
Top 0.1%
18.3%
Similar papers in this journal
2
Human Brain Mapping
○ 329 papers in training set
Top 0.5%
9.7%
Similar papers in this journal
3
Imaging Neuroscience
● 282 papers in training set
Top 0.8%
6.6%
Similar papers in this journal
4
IEEE Transactions on Medical Imaging
● 21 papers in training set
Top 0.1%
6.2%
5
Nature Communications
○ 5641 papers in training set
Top 25%
6.2%
Similar papers in this journal
6
PLOS Computational Biology
● 1863 papers in training set
Top 6%
6.2%
Similar papers in this journal
50% of probability mass above
7
NeuroImage
○ 903 papers in training set
Top 3%
5.4%
8
Scientific Reports
○ 3612 papers in training set
Top 23%
4.3%
9
PLOS ONE
● 5266 papers in training set
Top 34%
3.5%
10
Nature Machine Intelligence
○ 70 papers in training set
Top 1.0%
2.7%
11
Frontiers in Neuroscience
○ 256 papers in training set
Top 3%
2.1%
12
Patterns
○ 78 papers in training set
Top 1%
2.1%
13
Frontiers in Computational Neuroscience
○ 60 papers in training set
Top 0.7%
1.9%
14
Frontiers in Psychiatry
○ 87 papers in training set
Top 1%
1.9%
15
Scientific Data
○ 209 papers in training set
Top 1%
1.9%
16
IEEE Transactions on Biomedical Engineering
● 40 papers in training set
Top 0.7%
1.4%
17
Nature Human Behaviour
○ 95 papers in training set
Top 2%
1.1%
18
Nature
○ 645 papers in training set
Top 9%
1.1%
19
Frontiers in Neuroinformatics
○ 41 papers in training set
Top 0.7%
0.8%
20
Nature Methods
○ 385 papers in training set
Top 7%
0.8%
21
Communications Biology
○ 993 papers in training set
Top 31%
0.8%
22
IEEE Access
● 35 papers in training set
Top 2%
0.6%
23
Neuroinformatics
○ 46 papers in training set
Top 1.0%
0.6%
24
Brain Informatics
○ 10 papers in training set
Top 0.3%
0.6%
25
eneuro
● 439 papers in training set
Top 8%
0.6%
26
Magnetic Resonance in Medicine
○ 85 papers in training set
Top 0.7%
0.6%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.