Brain Age Estimation on T2-FLAIR Scans for Application to Multiple Sclerosis.
Authors
Affiliations (27)
Affiliations (27)
- UCL Hawkes Institute, University College London, London, UK.
- Department of Neuroradiology, King's College Hospital, London, UK.
- Queen Square Multiple Sclerosis Centre, Department of Neuroinflammation, UCL Queen Square Institute of Neurology, London, UK.
- MS Center Amsterdam, Radiology and Nuclear Medicine, Vrije Universiteit Amsterdam, Amsterdam Neuroscience, Amsterdam UMC Location VUmc, Amsterdam, the Netherlands.
- Department of Advanced Biomedical Sciences, University of Naples "Federico II", Naples, Italy.
- Bayer Plc, Reading, UK.
- Department of Neuroradiology, Charité-Universitätsmedizin Berlin, Corporate Member of Freie Universität Berlin, Humboldt-Universität Zu Berlin, Berlin, Germany.
- Movement Disorders, Neurostimulation and Neuroimaging, University Medicine Mainz, Mainz, Germany.
- Department of Neurology, Oslo University Hospital, Oslo, Norway.
- Department of Psychology, University of Oslo, Oslo, Norway.
- Institute of Clinical Medicine, University of Oslo, Oslo, Norway.
- Nuffield Department of Clinical Neurosciences, Medical Sciences Division, University of Oxford, Oxford, UK.
- Department of Medicine, Surgery and Neuroscience, University of Siena, Siena, Italy.
- Section of Neuroradiology, Department of Radiology, Hospital Universitari Vall D'hebron, Barcelona, Spain.
- Department of Neurology, Medical University of Graz, Graz, Austria.
- Neuroimaging Research Unit, Division of Neuroscience, IRCCS San Raffaele Scientific Institute, Vita-Salute San Raffaele University, Milan, Italy.
- Vita-Salute San Raffaele University, Milan, Italy.
- Neurology Unit, Neurorehabilitation Unit, Neurophysiology Service, and Neuroimaging Research Unit, Division of Neuroscience, IRCCS San Raffaele Scientific Institute, Milan, Italy.
- Institute of Neuroradiology, St. Josef Hospital, Ruhr-University Bochum, Bochum, Germany.
- Department of Neurology, St. Josef Hospital, Ruhr University Bochum, Bochum, Germany.
- Department of Neurosciences, Biomedicine and Movement Sciences, University of Verona, Verona, Italy.
- Department of Information Engineering, University of Padova, Padova, Italy.
- Department of Neurology and Center of Clinical Neuroscience, First Faculty of Medicine, Charles University and General University Hospital, Prague, Czech Republic.
- Department of Radiology, First Faculty of Medicine, Charles University and General University Hospital, Prague, Czech Republic.
- Lysholm Department of Neuroradiology, UCLH National Hospital for Neurology and Neurosurgery, London, UK.
- E-Health Center, Universitat Oberta de Catalunya, Barcelona, Spain.
- Dementia Research Centre, UCL Queen Square Institute of Neurology, London, UK.
Abstract
The brain-predicted age difference (brain-PAD) is associated with measures of clinical interest in people with multiple sclerosis (pwMS). Most brain age models rely on 3D T1-weighted scans, which are not routinely acquired in MS clinical practice, limiting their potential for clinical translation. We aimed to develop a model predicting brain age using T2-FLAIR, the core sequence for MS diagnosis and monitoring, and validate the resulting brain-PAD values as a biomarker of MS severity and progression. We collected 3D T2-FLAIR and 3D T1-weighted brain MRI scans to compose (i) a multicentre cohort of healthy participants for brain age modeling, and (ii) a single-centre cohort of pwMS and healthy controls for external validation. We trained and evaluated 3D convolutional neural network models predicting brain age from T2-FLAIR or T1-weighted images. Models were compared using t-tests based on bootstrapped standard errors. Saliency maps were obtained with the SmoothGrad method to visualize regions that were most important for the predictions. Finally, using a linear model framework, we clinically validated the resulting brain-PAD metric by assessing its relationship with diagnosis (MS versus healthy controls), clinical phenotype, disease duration, and physical disability as measured with the Expanded Disability Status Scale (EDSS), adjusting for age and sex. The Inception-ResNet-V2 model based on T2-FLAIR scans yielded accurate brain age predictions (test set MAE = 3.31 years, R<sup>2</sup> = 0.944, 5x ensemble MAE = 2.81, R<sup>2</sup> = 0.955), which were comparable to those obtained with the T1w-based model (test set MAE = 3.34 years, R<sup>2</sup> = 0.942, 5x ensemble MAE = 2.84, R<sup>2</sup> = 0.955, p = 0.91). Brain age predictions were mostly driven by subcortical regions, particularly the thalamus. T2-FLAIR-based brain-PAD was higher in pwMS than healthy controls (7.07 vs -0.50 years, p < 0.0001). As with T1 brain-PAD, FLAIR brain-PAD correlated with MS disease duration (R = 0.24, p < 0.0001) and EDSS (R = 0.30, p < 0.0001). Brain age predictions relying on T2-FLAIR scans are as accurate as those derived from T1-weighted scans and could be used as an easily obtainable biomarker of MS severity and progression in clinical practice.