Brain Age Gap Estimation (BrainAGE): Influence of scanner manufacturer, field strength and sequence
Poster No:
1164
Submission Type:
Abstract Submission
Authors:
Vivien Lorena Ivan1, Marius Vach1, Daniel Weiß1, Luisa Wolf1, Shammi More2,3, Kaustubh Patil2,3, Dennis Hedderich4,5, Nora Bittner6, Julian Caspers1, Christian Rubbert1
Institutions:
1Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital, Düsseldorf, NRW, Germany, 2Institute of Neuroscience and Medicine, Brain & Behaviour (INM-7), Research Centre Jülich, Jülich, NRW, Germany, 3Institute of Systems Neuroscience, Medical Faculty, Düsseldorf, NRW, Germany, 4Department of Neuroradiology, School of Medicine, Technical University of Munich, Munich, Bavaria, Germany, 5TUM-NIC Neuroimaging Center, School of Medicine, Technical University of Munich, Munich, Bavaria, Germany, 6Institute for Anatomy I, Medical Faculty, Heinrich-Heine University, Düsseldorf, NRW, Germany
First Author:
Vivien Lorena Ivan
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Co-Author(s):
Marius Vach
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Daniel Weiß
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Luisa Wolf
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Shammi More
Institute of Neuroscience and Medicine, Brain & Behaviour (INM-7), Research Centre Jülich|Institute of Systems Neuroscience, Medical Faculty
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Institute of Neuroscience and Medicine, Brain & Behaviour (INM-7), Research Centre Jülich|Institute of Systems Neuroscience, Medical Faculty
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Kaustubh Patil
Institute of Neuroscience and Medicine, Brain & Behaviour (INM-7), Research Centre Jülich|Institute of Systems Neuroscience, Medical Faculty
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Institute of Neuroscience and Medicine, Brain & Behaviour (INM-7), Research Centre Jülich|Institute of Systems Neuroscience, Medical Faculty
Jülich, NRW, Germany|Düsseldorf, NRW, Germany
Dennis Hedderich
Department of Neuroradiology, School of Medicine, Technical University of Munich|TUM-NIC Neuroimaging Center, School of Medicine, Technical University of Munich
Munich, Bavaria, Germany|Munich, Bavaria, Germany
Department of Neuroradiology, School of Medicine, Technical University of Munich|TUM-NIC Neuroimaging Center, School of Medicine, Technical University of Munich
Munich, Bavaria, Germany|Munich, Bavaria, Germany
Nora Bittner
Institute for Anatomy I, Medical Faculty, Heinrich-Heine University
Düsseldorf, NRW, Germany
Institute for Anatomy I, Medical Faculty, Heinrich-Heine University
Düsseldorf, NRW, Germany
Julian Caspers
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Christian Rubbert
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Department of Diagnostic and Interventional Radiology, Medical Faculty and University Hospital
Düsseldorf, NRW, Germany
Introduction:
Brain Age Gap Estimation (BrainAGE) is explored as an imaging biomarker for several conditions, such as neurodegenerative diseases [Gaser 2013], or in lifestyle-choices-driven aging [Bittner 2021]. BrainAGE is usually derived from applying machine-learning approaches to features derived from structural magnetic resonance images of the brain [Franke 2019].
In cognitively normal, healthy subjects a BrainAGE of zero is expected, i.e. the predicted and the chronological age are the same, which is the underlying assumption of the current study. However, differences in BrainAGE, especially outside of the training cohorts, could arise due to different combinations of, e.g., study in- and exclusion criteria, MRI scanners, field strength or sequence parameters. Thus, this study aims to systematically assess the influence of the aforementioned.
In cognitively normal, healthy subjects a BrainAGE of zero is expected, i.e. the predicted and the chronological age are the same, which is the underlying assumption of the current study. However, differences in BrainAGE, especially outside of the training cohorts, could arise due to different combinations of, e.g., study in- and exclusion criteria, MRI scanners, field strength or sequence parameters. Thus, this study aims to systematically assess the influence of the aforementioned.
Methods:
A total of 2,414 normal participants from four population based studies were included, specifically from ADNI (n = 914, 55% female, 72.8±6.4 years, 55-90 years), HCPA (n = 725, 56% female, 60.3±15.7 years, 36-100 years), OASIS3 (n = 609, 58.9% female, 67.8±8.9 years, 42-95 years) and PPMI (n = 166, 36.7% female, 60.5±11.6 years, 30-82 years).
BrainAGE was derived for each subject using the best-performing model from [More 2023], trained on 2,953 healthy controls (18-88 years, 3T scans only) from four population-based studies not used in the current study. In essence, gray matter features were derived from standardized CAT12 for SPM12 preprocessing, smoothing with a 4 mm full-width-half-maximum kernel, resampling to 4 mm spatial resolution, and finally by principal component analysis for dimensionality reduction. Finally, Gaussian process regression was used to predict age, and BrainAGE was calculated by subtracting the chronological from the predicted age.
BrainAGE was then compared across the different cohorts. Furthermore, differences in BrainAGE based on scanner manufacturer, field strength, and choice of unaccelerated or accelerated T1-imaging were analyzed for each cohort, when data was available. Statistically significant differences (p<0.05), were analyzed using Welch two sample t-test or ANOVA, including the Tukey post-hoc test, as applicable. The mean absolute error between chronological and predicted age was calculated.
BrainAGE was derived for each subject using the best-performing model from [More 2023], trained on 2,953 healthy controls (18-88 years, 3T scans only) from four population-based studies not used in the current study. In essence, gray matter features were derived from standardized CAT12 for SPM12 preprocessing, smoothing with a 4 mm full-width-half-maximum kernel, resampling to 4 mm spatial resolution, and finally by principal component analysis for dimensionality reduction. Finally, Gaussian process regression was used to predict age, and BrainAGE was calculated by subtracting the chronological from the predicted age.
BrainAGE was then compared across the different cohorts. Furthermore, differences in BrainAGE based on scanner manufacturer, field strength, and choice of unaccelerated or accelerated T1-imaging were analyzed for each cohort, when data was available. Statistically significant differences (p<0.05), were analyzed using Welch two sample t-test or ANOVA, including the Tukey post-hoc test, as applicable. The mean absolute error between chronological and predicted age was calculated.
Results:
Across the different cohorts, the BrainAGE for ADNI was -5.9±5.5 years (MAE 6.7), for HCPA -4.1±6.2 (MAE 5.9), for OASIS3 -4.8±5.4 (MAE 5.8), and for PPMI -3±5.8 (MAE 5.3) with a p<0.0001 (ANOVA) (Figure 1A). Only OASIS3/HCPA (p=0.17) and PPMI/HCPA (p=0.1) did not show significant differences.
Comparing the field strength (Figure 1B), we found that scans performed at 1.5T resulted in a smaller BrainAGE (ADNI: -2.6±4.9 years, MAE 4.2; OASIS3: -2.4±4.6, MAE 4.1, PPMI: -1.6±4.8, MAE 4.2) than scans acquired at 3T (ADNI: -6.9±5.3 years, MAE 7.5; OASIS3: -4.9±5.5, MAE 5.9, PPMI: -3.5±6.1, MAE 5.8). p<0.0001 for ADNI, p=0.007 for OASIS3, and p=0.03 for PPMI. HCPA only acquired scans at 3T.
Scans in ADNI and PPMI were acquired on scanners from different vendors (Figure 2A). In ADNI, scans acquired on GE scanners resulted in a BrainAGE of -6.3±6.2 years (MAE 7.3), on Philips of -4.9±5.1 years (MAE 6), and on Siemens of -6.1±5.3 years (MAE 6.7). p=0.03 (ANOVA), with statistically significant differences between Philips/GE (p=0.008) in the post-hoc analysis. For PPMI, the results were -0.5±7.7 years for GE scanners (MAE 6.2), -1.8±6.2 for Philips (MAE 5.1), and -4.1±4.8 for Siemens (MAE 5.1). p=0.005 (ANOVA), with statistically significant difference between Siemens/GE (p=0.008) in the post-hoc analysis.
In ADNI, accelerated (-7±5.3 years, MAE 7.6) and unaccelerated (-7±5.3, MAE 7.3) imaging was available acquired at 3T (p=0.47, Figure 2B).
Comparing the field strength (Figure 1B), we found that scans performed at 1.5T resulted in a smaller BrainAGE (ADNI: -2.6±4.9 years, MAE 4.2; OASIS3: -2.4±4.6, MAE 4.1, PPMI: -1.6±4.8, MAE 4.2) than scans acquired at 3T (ADNI: -6.9±5.3 years, MAE 7.5; OASIS3: -4.9±5.5, MAE 5.9, PPMI: -3.5±6.1, MAE 5.8). p<0.0001 for ADNI, p=0.007 for OASIS3, and p=0.03 for PPMI. HCPA only acquired scans at 3T.
Scans in ADNI and PPMI were acquired on scanners from different vendors (Figure 2A). In ADNI, scans acquired on GE scanners resulted in a BrainAGE of -6.3±6.2 years (MAE 7.3), on Philips of -4.9±5.1 years (MAE 6), and on Siemens of -6.1±5.3 years (MAE 6.7). p=0.03 (ANOVA), with statistically significant differences between Philips/GE (p=0.008) in the post-hoc analysis. For PPMI, the results were -0.5±7.7 years for GE scanners (MAE 6.2), -1.8±6.2 for Philips (MAE 5.1), and -4.1±4.8 for Siemens (MAE 5.1). p=0.005 (ANOVA), with statistically significant difference between Siemens/GE (p=0.008) in the post-hoc analysis.
In ADNI, accelerated (-7±5.3 years, MAE 7.6) and unaccelerated (-7±5.3, MAE 7.3) imaging was available acquired at 3T (p=0.47, Figure 2B).
·Figure 1
·Figure 2
Conclusions:
Researchers and clinicians should be aware that several factors, such as choice of scanner manufacturer and field strength, may influence certain Brain Age Gap Estimation (BrainAGE) and that comparisons across cohorts may not be suitable.
Lifespan Development:
Aging 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Other Methods
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Aging
Computing
Data analysis
Machine Learning
STRUCTURAL MRI
Other - BrainAGE
1|2Indicates the priority used for review
Provide references using author date format
Franke K, Gaser C (2019): Ten Years of BrainAGE as a Neuroimaging Biomarker of Brain Aging: What Insights Have We Gained? Front Neurol 10:789.
Gaser C, Franke K, Klöppel S, Koutsouleris N, Sauer H, Initiative ADN (2013): BrainAGE in Mild Cognitive Impaired Patients: Predicting the Conversion to Alzheimer’s Disease. PLoS ONE 8:e67346.
More S, Antonopoulos G, Hoffstaedter F, Caspers J, Eickhoff SB, Patil KR, Initiative ADN (2023): Brain-age prediction: A systematic comparison of machine learning workflows. NeuroImage 270:119947.