Back

Can deep learning predict human intelligence from structural brain MRI?

Hussain, M. A.; LaMay, D.; Grant, E.; Ou, Y.

2023-02-27 bioengineering
10.1101/2023.02.24.529924 bioRxiv
Show abstract

Can brain structure predict human intelligence? T1-weighted structural brain magnetic resonance images (sMRI) have been correlated with intelligence. Nevertheless, population-level association does not fully account for individual variability in intelligence. To address this, individual prediction studies emerge recently. However, they are mostly on predicting fluid intelligence (the ability to solve new problems). Studies are lacking to predict crystallized intelligence (the ability to accumulate knowledge) or general intelligence (fluid and crystallized intelligence combined). This study tests whether deep learning of sMRI can predict an individual subjects verbal, comprehensive, and full-scale intelligence quotients (VIQ, PIQ, FSIQ), which reflect both fluid and crystallized intelligence. We performed a comprehensive set of 432 experiments, using different input images, six deep learning models, and two outcome settings, on 850 autistic and healthy subjects 6-64 years of age. Results show promise with statistical significance, and also open up questions inviting further future studies.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

1
Scientific Reports
○ 3612 papers in training set
Top 0.7%
18.8%
Similar papers in this journal
2
Human Brain Mapping
○ 329 papers in training set
Top 0.1%
17.2%
Similar papers in this journal
3
Scientific Data
○ 209 papers in training set
Top 0.4%
6.3%
Similar papers in this journal
4
NeuroImage
○ 903 papers in training set
Top 2%
6.3%
Similar papers in this journal
5
Medical Image Analysis
○ 35 papers in training set
Top 0.2%
4.9%
Similar papers in this journal
50% of probability mass above
6
Nature Communications
○ 5641 papers in training set
Top 32%
4.1%
7
Communications Biology
○ 993 papers in training set
Top 4%
3.6%
8
Frontiers in Neuroscience
○ 256 papers in training set
Top 2%
2.4%
9
Computers in Biology and Medicine
○ 128 papers in training set
Top 2%
2.1%
10
Science Bulletin
○ 21 papers in training set
Top 0.1%
2.1%
11
PLOS ONE
● 5266 papers in training set
Top 42%
2.1%
12
Nature Machine Intelligence
○ 70 papers in training set
Top 1%
1.8%
13
Imaging Neuroscience
● 282 papers in training set
Top 3%
1.8%
14
iScience
○ 1154 papers in training set
Top 23%
1.4%
15
Bioengineering
○ 29 papers in training set
Top 0.6%
1.1%
16
eLife
● 5828 papers in training set
Top 58%
1.1%
17
Science Advances
● 1243 papers in training set
Top 24%
1.1%
18
IEEE Transactions on Medical Imaging
● 21 papers in training set
Top 0.3%
1.1%
19
PLOS Computational Biology
● 1863 papers in training set
Top 16%
1.1%
20
Neuroinformatics
○ 46 papers in training set
Top 0.6%
1.1%
21
Advanced Science
○ 286 papers in training set
Top 8%
0.9%
22
npj Digital Medicine
○ 118 papers in training set
Top 3%
0.9%
23
IEEE Access
● 35 papers in training set
Top 1%
0.9%
24
Brain Research Bulletin
○ 10 papers in training set
Top 0.1%
0.9%
25
Nature Methods
○ 385 papers in training set
Top 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.