Beyond Brain Age: A Density-Estimation Approach to Brain Structural Normativity
Poster No:
1919
Submission Type:
Abstract Submission
Authors:
Ramona Leenings1, Jan Ernsting2, Nils Winter3, Maximilian Konowski4, Lukas Fisch3, Daniel Emden3, Carlotta Barkhau3, Xiaoyi Jiang5, Udo Dannlowski3, Klaus Berger6, Tim Hahn3
Institutions:
1University of Münster, Münster, North Rhine Westphalia, 2University of Münster, Münster, NRW, 3Institute for Translational Psychiatry, Münster, North Rhine Westphalia, 4University of Münster, Münster, North-Rhine Westphalia, 5University of Münster, Münster, Northrine-Westphaila, 6Institute of Epidemiology and Social Medicine, University of Münster, Münster, NRW, Germany
First Author:
Co-Author(s):
Klaus Berger
Institute of Epidemiology and Social Medicine, University of Münster
Münster, NRW, Germany
Institute of Epidemiology and Social Medicine, University of Münster
Münster, NRW, Germany
Introduction:
In recent years, Brain Age models have been used to quantify neuro-structural degeneration and the resulting biomarker, the Brain Age Gap (BAG), has been linked to numerous neurological and psychiatric conditions (Cole et al. 2019, Bittner et al. 2021, Wrigglesworth et al. 2021, Blake et al. 2023). The conventional Brain Age concept assumes the existence of discernible, one-year increments of neurostructural changes along the age continuum. In practice, however, a brain age model needs to resolve the conflict of neuro-structurally similar brains with differing age labels and vice versa. In addition, the concept neglects inter-individual variety and implicitly assumes that neurostructural irregularities within one age group are normative in another. We therefore propose to reformulate the learning problem from "how old is this brain?" to "how unusual is this brain structure given its chronological age?". This approach explicitly allows for several prototypical brain structures per age group and overlapping prototypes across several (neighboring) age groups. In addition, this approach offers individual normativity scores for all age groups along the aging continuum, thereby generating an individual aging profile (see Figure 1). We introduce two novel biomarkers termed 'Group Normativity', which describes age-group specific brain structural normativity, and 'Profile Normativity', which describes the normativity of the aging profile with respect to the aging profiles of a same-aged representative sample.
Methods:
We utilize a k-Nearest-Neighbor algorithm to estimate the typical regional density for a representative sample of a particular age group. Subsequently, we aggregate the normalized distances to the k=15 nearest neighbors for all training samples, fit a distribution, and derive a normativity score from this process. A total of N=30,047 participants from the German National Cohort (German National Consortium 2014, Peters et al. 2022) was used to derive models for age groups 21 to 72. For simplicity, we describe the brain structures using five global tissue variables (GM, WM, WMH, CSF and TIV) derived with the cat12 toolbox (Gaser, https://neuro-jena.github.io/cat/). To assess detection capabilities for atypical neurostructural degeneration, we evaluate our biomarkers using instances of Mild Cognitive Impairment (MCI) and Alzheimer's Disease (AD) from the ADNI dataset. We compare our novel biomarkers against two Brain Age benchmarks: a SVM-based Brain Age model (BAG-SVM) using the same five variables and a deep learning-based Brain Age model (BAG-DL) trained on full structural T1 images (Hahn et al. 2022, Ernsting et al. 2023). Utilizing a machine learning pipeline implemented in the Python package photonai (Leenings et al. 2021), we classify control subjects (n=351) from those with MCI (n=338) or AD (n=163) with each of the biomarkers.
·Actual age (green) and the normativity (blue) of a particular brain structure compared to representative samples of the different age groups.
Results:
Both novel biomarkers are able to measure atypical neurostructural degeneration: The Profile Normativity score demonstrates robust performance in distinguishing Alzheimer's Disease (AD) and Mild Cognitive Impairment (MCI), showcasing notably high recall rates of 0.86 for AD and 0.67 for MCI, surpassing BAG-SVM and BAG-DL (see Figure 1). For AD, both novel biomarkers outperform the best Brain Age model (BAG-DL) in terms of balanced accuracy (0.73 and 0.82 vs 0.71) and f1 score (0.64 and 0.7 vs. 0.61). For MCI, the Group Normativity biomarker seems to lack informative value, while the Profile Normativity biomarker is able to compete with the best Brain Age model (BAG-SVM) in terms of balanced accuracy (0.64 to 0.65) and outperforms the f1 score (0.65 to 0.61).
·Benchmarking the detection of atypical neurostructural degeneration for each biomarker.
Conclusions:
By focusing on structural deviations relative to chronological age, our approach accommodates the variability in healthy aging trajectories that the current models overlook. The here presented approach was able to compete with or outperform large-scale deep neural networks and yields two promising biomarkers for future scientific pursuits.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s)
Lifespan Development:
Aging 2
Modeling and Analysis Methods:
Classification and Predictive Modeling
Methods Development 1
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Aging
Machine Learning
Modeling
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
Blake, K. V. et al. Advanced brain ageing in adult psychopathology: A systematic review and meta-analysis of structural MRI studies. J. Psychiatr. Res. 157, 180–191 (2023).
Cole, J. H., Marioni, R. E., Harris, S. E. & Deary, I. J. Brain Age and Other Bodily ‘Ages’: Implications for Neuropsychiatry. Mol Psychiatr 24, 266–281 (2019).
Ernsting, J. et al. From Group-Differences to Single-Subject Probability: Conformal Prediction-based Uncertainty Estimation for Brain-Age Modeling. arXiv (2023)
doi:10.48550/arxiv.2302.05304.
German National Consortium (GNC). The German National Cohort: aims, study design and organization. Eur. J. Epidemiology 29, 371–382 (2014).
Hahn, T. et al. An Uncertainty-Aware, Shareable, and Transparent Neural Network Architecture for Brain-Age Modeling. Sci Adv 8, eabg9471 (2022).
Leenings, R. et al. PHOTONAI—A Python API for Rapid Machine Learning Model Development. Plos One 16, e0254062 (2021).
Peters, A. et al. Framework and baseline examination of the German National Cohort (NAKO). Eur. J. Epidemiology 37, 1107–1124 (2022).
Wrigglesworth, J. et al. Factors associated with brain ageing - a systematic review. BMC Neurol. 21, 312 (2021)