Assessing highly accelerated 3D-T1w Wave CAIPI MPRAGE images for brain age prediction in dementia
Poster No:
2294
Submission Type:
Abstract Submission
Authors:
Rafael Navarro-González1, Santiago Aja-Fernández1, Rodrigo de Luis Garcia1, Daniel Alexander2, Frederik Barkhof2,3, Millie Beament3, Haroon Chughtai2,4, Nick Fox3, Catherine Mummery3, Miguel Rosa-Grilo3, David Thomas3, Geoff Parker2, James Cole2,3
Institutions:
1Universidad de Valladolid, Valladolid, Spain, 2Centre for Medical Image Computing, UCL, London, United Kingdom, 3Dementia Research Centre, UCL, London, United Kingdom, 4Centre for Advanced Research Computing, UCL, London, United Kingdom
First Author:
Co-Author(s):
Frederik Barkhof, MD, Ph. D
Centre for Medical Image Computing, UCL|Dementia Research Centre, UCL
London, United Kingdom|London, United Kingdom
Centre for Medical Image Computing, UCL|Dementia Research Centre, UCL
London, United Kingdom|London, United Kingdom
Haroon Chughtai
Centre for Medical Image Computing, UCL|Centre for Advanced Research Computing, UCL
London, United Kingdom|London, United Kingdom
Centre for Medical Image Computing, UCL|Centre for Advanced Research Computing, UCL
London, United Kingdom|London, United Kingdom
James Cole, PhD
Centre for Medical Image Computing, UCL|Dementia Research Centre, UCL
London, United Kingdom|London, United Kingdom
Centre for Medical Image Computing, UCL|Dementia Research Centre, UCL
London, United Kingdom|London, United Kingdom
Introduction:
Traditional MRI sequences, such as the standard T1-weighted Magnetization Prepared Rapid Gradient Echo (MPRAGE), have been the cornerstone for neuroimaging. However, the advent of advanced accelerated imaging techniques like Wave Controlled Aliasing In Parallel Imaging (Wave CAIPI), results in shorter scan times, presenting a shift in data acquisition speed [1].
Brain age is a measure derived from neuroimaging that estimates the brain's 'biological age', calculated by applying machine learning methods [2]. The disparity between predicted age and real age, known as the brain-predicted age difference (BrainPAD), aims to reflect differences in brain aging, which can be influenced by lifestyle, genetics, and disease [3-5]. This work compares brain age predictions derived from accelerated Wave CAIPI MPRAGE acquisitions against those obtained from standard MPRAGE acquisitions.
Brain age is a measure derived from neuroimaging that estimates the brain's 'biological age', calculated by applying machine learning methods [2]. The disparity between predicted age and real age, known as the brain-predicted age difference (BrainPAD), aims to reflect differences in brain aging, which can be influenced by lifestyle, genetics, and disease [3-5]. This work compares brain age predictions derived from accelerated Wave CAIPI MPRAGE acquisitions against those obtained from standard MPRAGE acquisitions.
Methods:
We studied 124 individuals from the Biomarkers and Rapid Imaging in Dementia Diagnosis (B-RAPIDD) dataset [6]. Participants were patients from the Cognitive Disorders Clinics at the National Hospital for Neurology and Neurosurgery (London, United Kingdom) with different diagnoses, undergoing investigations for memory and cognitive difficulties. Participants undertook a standard 312 seconds T1w MPRAGE acquisition following the Alzheimer's Disease Neuroimaging Initiative 2 protocol [7], and one or two MPRAGE sequences with Wave CAIPI undersampling. Additionally, all individuals received the mini-mental state exam (MMSE). Acquisition parameters are comparable among the Wave CAIPI acquisitions so for each participant with a standard acquisition an accelerated one was randomly selected. For details of acquisition parameters and demographics, see Figure 1.
The brainageR model [8] was used to estimate brain age from the standard and accelerated scans. Performance was evaluated using the mean absolute error (MAE), Pearson's correlation (r), and R-squared (R2). High MAE was expected since cognitively impaired subjects were assessed. Reliability and agreement between scan types were analyzed using the intra-class correlation coefficient (ICC) and Pearson's correlation. The brain age difference between standard and Wave CAIPI scans was assessed, along with a t-test to determine its significance. Additionally, the influence of scan type on cognitive status was explored by correlating BrainPADs with MMSE scores, with age and sex as covariates. The significance of the difference between the calculated R2 and the correlation coefficients was assessed using bootstrapping with 1,000 iterations.
The brainageR model [8] was used to estimate brain age from the standard and accelerated scans. Performance was evaluated using the mean absolute error (MAE), Pearson's correlation (r), and R-squared (R2). High MAE was expected since cognitively impaired subjects were assessed. Reliability and agreement between scan types were analyzed using the intra-class correlation coefficient (ICC) and Pearson's correlation. The brain age difference between standard and Wave CAIPI scans was assessed, along with a t-test to determine its significance. Additionally, the influence of scan type on cognitive status was explored by correlating BrainPADs with MMSE scores, with age and sex as covariates. The significance of the difference between the calculated R2 and the correlation coefficients was assessed using bootstrapping with 1,000 iterations.
Results:
Both standard and Wave CAIPI acquisitions achieve similar predictions of brain age (see Figure 2), with comparable performance in terms of their MAE (standard MAE = 8.16 years; Wave CAIPI MAE = 7.79 years) and correlation (standard r = 0.83; Wave CAIPI r = 0.82). Additionally, the correlation between them, including intra-class and Pearson correlation (ICC = 0.95, 95% CI [0.93, 0.97]; r = 0.96), was very high. However, with Wave CAIPI, there was a tendency to underestimate age in younger subjects (Figure 2C). Further, the Wave CAIPI method seems to exhibit a higher correlation with MMSE (Standard r = -0.25; Wave CAIPI r = -0.31; difference = 0.07 IC 95 % [-0.21, 0.38]), although neither these or the R2 (Standard R2 = 0.15; Wave CAIPI R2 = 0.18; difference -0.03 IC 95 % [-0.19, 0.11]) were significantly different (Figure 2D).
Conclusions:
These observations suggest that the accelerated technique efficiently shortens data acquisition time without compromising the accuracy of predictions. It is important to recognize that the biases observed in age estimation are likely due to subtle differences in image characteristics relative to the brain age model training data, and do not necessarily reflect flaws in the accelerated method. Furthermore, the sensitivity of the Wave CAIPI method to cognitive changes appears to be better, but not significantly so.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Novel Imaging Acquisition Methods:
Anatomical MRI 1
Keywords:
Aging
Machine Learning
MRI
Other - Brain Age
1|2Indicates the priority used for review
Provide references using author date format
[2] Franke, K. (2010), "Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: exploring the influence of various parameters." Neuroimage 50.3 883-892.
[3] Cole, J. H. (2020), "Multimodality neuroimaging brain-age in UK biobank: relationship to biomedical, lifestyle, and cognitive factors." Neurobiology of aging 92 34-42.
[4] Leonardsen, E. H. (2023), "Genetic architecture of brain age and its casual relations with brain and mental disorders." Molecular Psychiatry 1-10.
[5] Cole, J. H. (2020), "Longitudinal assessment of multiple sclerosis with the brain‐age paradigm." Annals of neurology 88.1 93-105.
[6] NHS Choices, NHS, www.hra.nhs.uk/planning-and-improving-research/application-summaries/research-summaries /biomarkers-and-rapid-imaging-in-dementia-diagnosis-b-rapidd/. Accessed 7 Nov. 2023.
[7] Clifford J. R. (2015) "Magnetic resonance imaging in Alzheimer's disease neuroimaging initiative 2." Alzheimer's & Dementia 11.7 740-756.
[8] Cole, J. H. (2018), "Brain age predicts mortality." Molecular psychiatry 23.5 1385-1392.
Acknowledgements
This work was funded by the Alzheimer’s Society Heather Corrie Impact Fund (grant number 577 [AS-PG-21-045]) and Biogen Idec UK. This work was supported by Ministerio de Ciencia e Innovación of Spain with research grants PID2021-124407NB-I00, TED2021-130758B-I00 and PRE2019-089176.