Brain-age in ultra-low-field MRI compared to high-field MRI
Poster No:
2307
Submission Type:
Abstract Submission
Authors:
Francesca Biondo1, Carly Bennalick2, Sophie Martin1, František Váša2, James Cole1
Institutions:
1University College London, United Kingdom, 2King's College London, United Kingdom
First Author:
Co-Author(s):
Introduction:
Recent advancements in portable ultra-low-field (ULF) MRI systems, defined as 50-100mT, provide an opportunity for broader neuroimaging applications, including in low and middle-income countries. These systems promise lower scanning costs and wider global adoption, addressing the limitations of high cost and infrastructural requirements of traditional MRI systems. This study assesses the accuracy of brain-age predictions from ULF MRI compared to high-field (HF) 3T MRI. Brain-age is an estimate of age derived from a model trained to predict age from neuroimaging features in a cohort of healthy adults. This cohort establishes a norm for age prediction and thus, brain-age can serve as an index of neurobiological health (Cole & Franke, 2017).
Methods:
Twenty-three healthy adults (mean age (SD) = 45.1 (15.3) years, range = 21.0-69.0 years; 12 males) were each scanned using a 3T MRI HF scanner (T1 MPRAGE 1 mm isotropic) and two identical Hyperfine Swoop ULF scanners at 64mT (three orthogonal T1 acquisitions with high-resolution in the Axial / Coronal / Sagittal plane; 1.6 x 1.6 x 5 mm). The data were acquired by the Department of Neuroimaging at King's College London and Guy's and St. Thomas' Hospital, London. For further details on all authors involved in data acquisition, refer to OHBM abstract 'Portable ultra-low-field brain MRI: test-retest reliability and correspondence to high-field MRI'. Images from the ULF scanners were linearly combined into a single T1 scan using multi-resolution registration (MRR; Deoni et al., 2022). Brain-age was then estimated for each MRR scan using the brainageR model, trained to predict age from an independent set of 3377 HF T1-weighted MRI scans of healthy adults across different sites (www.github.com/james-cole/brainageR).
Results:
The HF scans demonstrated a strong correlation between brain-age and actual age (r = 0.79, R² = 0.02, MAE = 12.44). Conversely, the ULF scanners showed a weaker correlation: ULF1 (r = 0.33, R² = 0.08, MAE = 12.88) and ULF2 (r = 0.38, R²= 0.12, MAE = 12.44). The intercorrelation results between high-field (HF) 3T MRI scans and two ultra-low field (ULF) MRI scanners, ULF1 and ULF2, yielded Pearson correlation coefficients of r = 0.44 for HF vs ULF1, r = 0.50 for HF vs ULF2, and r = 0.90 for ULF1 vs ULF2 (across site, test-retest reliability).
Conclusions:
ULF MRI scanners, with their lower operational costs, broader reach to a wider section of the population and reduced infrastructural demands, offer a promising solution for expanding MRI accessibility, particularly in resource-limited settings. However, currently, they do not match the accuracy of HF 3T MRI in brain-age estimation. These discrepancies may largely stem from the generalisability limitations of the brainageR model. Indeed, the suboptimal brain-age estimates observed in the HF MRI data are likely to be amplified in the context of ULF data, considering brainageR was originally trained on HF data. The weak correlation between age and predicted age in ULF scans suggests a strong regression-to-the mean effect. Additionally, although the lower signal-to-noise ratio inherent to ULF imaging could be a factor, the good correspondence in grey matter volume estimates between HF and ULF (unpublished data) suggests it may not be the primary issue affecting brain-age accuracy. Notably, the intercorrelation between the brain-age estimates from the two ULF scanners was high suggesting high between-scanner test-retest reliability, despite the lower correlation with the HF estimates. Advancements such as super-resolution optimisation, inclusion of ULF T2 sequences, application of other brain-age models and use of transfer learning for dataset calibration could potentially mitigate this effect and improve the utility of ULF MRI for brain health assessments. These implementations are work-in-progress.
Lifespan Development:
Aging
Modeling and Analysis Methods:
Methods Development 2
Novel Imaging Acquisition Methods:
Anatomical MRI 1
Keywords:
Acquisition
Aging
STRUCTURAL MRI
Other - ultra-low-field MRI
1|2Indicates the priority used for review
Provide references using author date format
Deoni, S. C. L., O'Muircheartaigh, J., Ljungberg, E., Huentelman, M., & Williams, S. C. R. (2022). Simultaneous high-resolution T2-weighted imaging and quantitative T2 mapping at low magnetic field strengths using a multiple TE and multi-orientation acquisition approach. Magnetic Resonance in Medicine, 88(3), 1273-1281. doi:10.1002/mrm.29273