Poster No:
134
Submission Type:
Abstract Submission
Authors:
Jing Li1, Chiu Wa Lam1, Hanna Lu1
Institutions:
1The Chinese University of Hong Kong, Hong Kong, China
First Author:
Jing Li
The Chinese University of Hong Kong
Hong Kong, China
Co-Author(s):
Chiu Wa Lam
The Chinese University of Hong Kong
Hong Kong, China
Hanna Lu
The Chinese University of Hong Kong
Hong Kong, China
Introduction:
In the ever-evolving realms of neuroscience and cognitive health, the concept of 'brain-predicted age' has merged as a fascinating and enlightening paradigm. 'Brain-predicted age' harnesses advanced computational algorithms to analyse a wide range of neuroimaging data, facilitating early detection and prediction of dementia and cognitive disorders(Franke and Gaser 2019). However, in contrast to conventional statistical models, computational models often lack the capacity to offer neuroanatomical interpretability and specificity. In other words, most machine learning approaches typically fail to reveal ageing-related regional alterations in brain structure or their contributions to 'brain-predicted age'. This dearth of transparency presents significant limitations within the domain of 'brain-predicted age', particularly in clinical applications. To address this challenge, we have pioneered the implementation of mutual information (MI) to quantitatively assess, rank, and elucidate the distinct contributions and relevance of various cortical structures to 'brain-predicted age'.
Methods:
We developed a brain age prediction model utilizing the support vector regression (SVR) machine. For the training set, we employed T1-weighted MRI scans of 609 healthy participants (18-88 years of age), sourced from the Cam-CAN dataset. The testing set was comprised of 547 healthy subjects, aged 19.98-86.32 years, selected from the Brain-development (IXI) dataset. All T1-weighted MRIs underwent pre-processing and quantification using BrainSuite software into four distinct regional brain feature types: mean cortical thickness (GMT), gray matter volume (GMV), white matter volume (WMV), and cerebrospinal fluid (CSF) volume. The Kraskov (KSG) method was applied to calculate the MI score between each regional brain feature and 'brain-predicted age' within the testing set. In addition to calculating MI scores for the four individual brain feature types, we also computed MI scores for two unique combinations of these feature types. These included the assessment of regional brain parenchymal volume, which was a combination of GMV and WMV, as well as regional intracranial total volume, derived from the combination of GMV, WMV, and CSF volume.
Results:
After age-bias correction, the trained brain age prediction model exhibited the following performance metrics within the training set: MAE=5.15 years, RMSE=6.27 years, and R^2=0.88. In the testing set, where the model's performance was evaluated, the following results were observed: MAE=6.65 years, RMSE=8.53 years, and R^2=0.74. Among the four individual cortical features, GMV exhibited the most substantial total MI value (8.705), with the pre-central gyrus recording the highest MI score (0.694). The second-highest total MI value was associated with CSF volume (7.760), with the cingulate gyrus displaying the highest MI score (0.872). The third-highest total MI value corresponded to mean GMT (6.222), with the superior temporal gyrus achieving the highest MI value (0.526). Conversely, WMV demonstrated the lowest total MI value (4.594), with the highest MI value observed in the insula (0.349). In terms of brain parenchymal volume, the superior frontal gyrus exhibited the highest total MI value (0.804). In the context of intracranial total volume, the cingulate gyrus displayed the highest total MI value (1.181).
Conclusions:
We identified GMV emerged as the paramount influence in the determination of 'brain-predicted age', underscoring its pivotal role within the four distinct brain feature types in the context of age-related considerations. Moreover, the superior frontal gyrus and the cingulate gyrus displayed significant importance within the construct of 'brain-predicted age'
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Lifespan Development:
Aging
Lifespan Development Other
Modeling and Analysis Methods:
Classification and Predictive Modeling
Methods Development 2
Keywords:
Aging
Computational Neuroscience
Cortical Layers
Data analysis
Degenerative Disease
Machine Learning
Modeling
MRI
Open Data
1|2Indicates the priority used for review
Provide references using author date format
Franke, K. and C. Gaser (2019). "Ten years of BrainAGE as a neuroimaging biomarker of brain aging: what insights have we gained?" Frontiers in neurology: 789.