Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Editorial
. 2019 Aug 15:197:652-656.
doi: 10.1016/j.neuroimage.2018.10.003. Epub 2018 Oct 6.

Machine learning in neuroimaging: Progress and challenges

Christos Davatzikos  1
Affiliations
Editorial

Machine learning in neuroimaging: Progress and challenges

Christos Davatzikos. Neuroimage. .
No abstract available

PubMed Disclaimer

Figures

Fig. 1.
Fig. 1.
Publications obtained from Pubmed with the following search query: ("MRI” OR “Magnetic Resonance Imaging” OR "Structural Magnetic Resonance Imaging" OR "Functional Magnetic Resonance Imaging" OR "Diffusion Tensor Imaging" OR "FDG-PET” OR "Amyloid-PET” OR “multimodal" OR "neuroimaging") AND (“brain”) AND (“Machine learning” OR “pattern classification”), on September 5, 2018. This plot reflects the exponential growth of adoption of these methods in neuroimaging.
Fig. 2.
Fig. 2.
Different Brain-age regressors were fit to data of an aging population, using as features a simple set of approximately 150 ROI volumes parcelating the brain. All models gave relatively good cross-validated accuracy on cross-sectional datasets, with correlation coefficients varying from 0.8 to 0.84 (r = 0.84 was achieved via the linear SVR and the multi-layer perceptron artificial neural network using 5 hidden layers). However, when these models were applied prospectively to longitudinal data, only the linear SVR gave a distribution of rate of change of the brain-age scores centered around 1. Although ground truth is not available for these datasets, it is reasonable to assume that these brains age by approximately 1 year per year, plus/minus some range that defines accelerated/resilient brain aging. Of all these models, the simpler linear SVR regressor is therefore the best one, for this specific problem.
Fig. 3.
Fig. 3.
Volumetric measurements of the hippocampus from 13 different studies. (ROI volumes were obtained via an optimized multi-atlas, multi-warping consensus method (Doshi et al., 2016), which topped the challenge of (Asman, 2013)). Inter-study differences render it difficult to combine data into a single training set. Application of statistical harmonization methods (taking into consideration various covariates, such as age and sex, as well as the nonlinearity in brain aging trajectories) must be applied, prior to being able to leverage the power of such large databases for machine learning methods.
See this image and copyright information in PMC

References

    1. Asman A, 2013. In: MICCAI 2013 Challenge Workshop on Segmentation: Algorithms, Theory and Applications (SATA). MICCAI SATA 2013 competition]. Available from: http://masi.vuse.vanderbilt.edu/submission/leaderboard.html.
    1. BREIMAN L, 2001. Random forests. Mach. Learn 45, 5–32.
    1. Brodersen KH, et al., 2011. Generative embedding for model-based classification of fMRI data. PLoS Comput. Biol 7 (6), e1002079. - PMC - PubMed
    1. Bzdok D, et al., 2015. Semi-supervised factored logistic regression for high-dimensional neuroimaging data. In: NIPS 2015.
    1. Davatzikos C, 2004. Why voxel-based morphometric analysis should be used with great caution when characterizing group differences. Neuroimage 23 (1), 17–20. - PubMed

Publication types