Estimating age-related changes in in vivo cerebral magnetic resonance angiography using convolutional neural network

• Published in: Neurobiology of aging Vol. 87; pp. 125 - 131
• Main Authors: Nam, Yoonho, Jang, Jinhee, Lee, Hea Yon, Choi, Yangsean, Shin, Na Young, Ryu, Kang-Hyun, Kim, Dong Hyun, Jung, So-Lyung, Ahn, Kook-jin, Kim, Bum-soo
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.03.2020
• Subjects: Atherosclerosis; Cerebral arteries; Deep learning; Hypertension; MR angiography; Hypertension; Deep learning; Cerebral arteries; MR angiography; Atherosclerosis
• ISSN: 0197-4580; 1558-1497; 1558-1497
• DOI: 10.1016/j.neurobiolaging.2019.12.008

Although age-related changes of cerebral arteries were observed in in vivo magnetic resonance angiography (MRA), standard tools or methods measuring those changes were limited. In this study, we developed and evaluated a model to measure age-related changes in the cerebral arteries from 3D MRA using a 3D deep convolutional neural network. From participants without any medical abnormality, training (n = 800) and validation sets (n = 88) of 3D MRA were built. After preprocessing and data augmentation, a 3D convolutional neural network was trained to estimate each subject's chronological age from in vivo MRA data. There was good correlation between chronological age and predicted age (r = 0.83) in an independent test set (n = 354). The predicted age difference (PAD) of the test set was 2.41 ± 6.22. Interaction term between age and sex was significant for PAD (p = 0.008). After correcting for age and interaction term, men showed higher PAD (p < 0.001). Hypertension was associated with higher PAD with marginal significance (p = 0.073). We suggested that PAD might be a potential measurement of cerebral vascular aging. •There was good correlation between age at the time of MRI and predicted age from brain MRA (r = 0.83).•Using deep learning, we can assess the appropriateness of the overall cerebral vascular aging status of each subject.•This system can be used to assess cerebral vascular aging in subjects with atherosclerosis risk factors, such as hypertension.•A deep learning machine showed potential for assessing age-related changes with brain MRA without knowledge-based features.