Derivation of machine learning brain aging biomarkers for a set of forty thousand functional connectomes.
Authors
Affiliations (7)
Affiliations (7)
- Glenn Biggs Institute for Alzheimer's & Neurodegenerative Disease, University of Texas Health Science Center at San Antonio, 4940 Charles Katz Drive, San Antonio, 78229, TX, USA.
- Department of Epidemiology, University of Washington School of Public Health, 3980 15th Ave NE, Box 351621, Seattle, 98195, WA, USA.
- Section of Gerontology and Geriatric Medicine Department, Wake Forest University School of Medicine, Medical Center Boulevard, Winston-Salem, 27157, NC, USA.
- Glenn Biggs Institute for Alzheimer's & Neurodegenerative Disease, University of Texas Health Science Center at San Antonio, 4940 Charles Katz Drive, San Antonio, 78229, TX, USA; Department of Biostatistics, Boston University School of Public Health, 715 Albany Street, Boston, 02118, MA, USA.
- Department of Neurology, University of California Davis, 1651 Alhambra Blvd Suite 200A, Sacramento, 95816, CA, USA.
- Department of Biostatistics, Boston University School of Public Health, 715 Albany Street, Boston, 02118, MA, USA.
- Glenn Biggs Institute for Alzheimer's & Neurodegenerative Disease, University of Texas Health Science Center at San Antonio, 4940 Charles Katz Drive, San Antonio, 78229, TX, USA. Electronic address: [email protected].
Abstract
Various Magnetic Resonance Imaging modalities were developed to explore the brain. Among them, functional MRI is of key importance for studying brain activity and its neural substrates. Recent works have pointed out that machine learning can use neuroimaging data to predict brain age. This approach is crucial not only for understanding the effects of aging but also for refining diagnostics because many chronic and neurodegenerative diseases appear as accelerated aging. Unfortunately, the prediction of brain age is particularly challenging for functional data due to the large dimension of the high-resolution connectomes usually derived to summarize the functional organization of the brain and their particular mathematical properties. In this work, we investigate the prediction of brain age from functional data on a large scale by creating a set of forty thousand functional connectomes via the processing of the resting-state fMRI scans of four cohort studies. This dataset is used to explore the ability of various connectome transformations and machine learning strategies to achieve accurate age predictions. We hope that our results will open the way for more reliable functional brain age measures.