Benchmarking the generalizability of brain age models: Challenges posed by scanner variance and prediction bias

Human Brain Mapping, Journal Year: 2022, Volume and Issue: 44(3), P. 1118 - 1128

Published: Nov. 8, 2022

Abstract Machine learning has been increasingly applied to neuroimaging data predict age, deriving a personalized biomarker with potential clinical applications. The scientific and value of these models depends on their applicability independently acquired scans from diverse sources. Accordingly, we evaluated the generalizability two brain age that were trained across lifespan by applying them three distinct early‐life samples participants aged 8–22 years. These chosen based size diversity training data, but they also differed greatly in processing methods predictive algorithms. Specifically, one model was built gradient tree boosting (GTB) extracted features cortical thickness, surface area, volume. other 2D convolutional neural network (DBN) minimally preprocessed slices T1‐weighted scans. Additional variants created understand how changed when each became more similar test terms acquisition protocols. Our results illustrated numerous trade‐offs. GTB predictions relatively accurate overall yielded reliable lower quality In contrast, DBN displayed most utility detecting associations between gaps cognitive functioning. Broadly speaking, largest limitations affecting protocol differences biased estimates. If such confounds could eventually be removed without post‐hoc corrections, may have greater as biomarkers healthy aging.

Language: Английский

---

Brain‐wide associations between white matter and age highlight the role of fornix microstructure in brain ageing

Human Brain Mapping, Journal Year: 2023, Volume and Issue: 44(10), P. 4101 - 4119

Published: May 17, 2023

Unveiling the details of white matter (WM) maturation throughout ageing is a fundamental question for understanding brain. In an extensive comparison brain age predictions and age-associations WM features from different diffusion approaches, we analyzed UK Biobank magnetic resonance imaging (dMRI) data across midlife older (N = 35,749, 44.6-82.8 years age). Conventional advanced dMRI approaches were consistent in predicting age. WM-age associations indicate steady microstructure degeneration with increasing to ages. Brain was estimated best when combining showing aspects contributing Fornix found as central region complement forceps minor another important region. These regions exhibited general pattern positive intra axonal water fractions, axial, radial diffusivities, negative relationships mean fractional anisotropy, kurtosis. We encourage application multiple detailed insights into WM, further investigation fornix potential biomarkers ageing.

Language: Английский

---

A review on brain age prediction models

Brain Research, Journal Year: 2023, Volume and Issue: 1823, P. 148668 - 148668

Published: Nov. 10, 2023

Language: Английский

---

Comparison of Machine Learning Models for Brain Age Prediction Using Six Imaging Modalities on Middle-Aged and Older Adults

Sensors, Journal Year: 2023, Volume and Issue: 23(7), P. 3622 - 3622

Published: March 30, 2023

Machine learning (ML) has transformed neuroimaging research by enabling accurate predictions and feature extraction from large datasets. In this study, we investigate the application of six ML algorithms (Lasso, relevance vector regression, support extreme gradient boosting, category boost, multilayer perceptron) to predict brain age for middle-aged older adults, which is a crucial area in neuroimaging. Despite plethora proposed models, there no clear consensus on how achieve better performance prediction population. Our study stands out evaluating impact both image modalities using cohort cognitively normal adults aged 44.6 82.3 years old (N = 27,842) with modalities. We found that predictive more reliant used than employed. Specifically, our highlights superior T1-weighted MRI diffusion-weighted imaging demonstrates multi-modality-based significantly enhances compared unimodality. Moreover, identified Lasso as most algorithm predicting age, achieving lowest mean absolute error single-modality multi-modality predictions. Additionally, also ranked highest comprehensive evaluation relationship between BrainAGE five frequently mentioned BrainAGE-related factors. Notably, shows ensemble outperforms when computational efficiency not concern. Overall, provides valuable insights into development reliable models significant implications clinical practice research. findings highlight importance modality selection emphasize promising prediction.

Language: Английский

---

Genetic architecture of brain age and its causal relations with brain and mental disorders

Molecular Psychiatry, Journal Year: 2023, Volume and Issue: 28(7), P. 3111 - 3120

Published: May 10, 2023

Abstract The difference between chronological age and the apparent of brain estimated from imaging data—the gap (BAG)—is widely considered a general indicator health. Converging evidence supports that BAG is sensitive to an array genetic nongenetic traits diseases, yet few studies have examined architecture its corresponding causal relationships with common disorders. Here, we estimate using state-of-the-art neural networks trained on scans 53,542 individuals (age range 3–95 years). A genome-wide association analysis across 28,104 (40–84 years) UK Biobank revealed eight independent genomic regions significantly associated ( p < 5 × 10 −8 ) implicating neurological, metabolic, immunological pathways – among which seven are novel. No significant correlations or were found for Parkinson’s disease, major depressive disorder, schizophrenia, but two-sample Mendelian randomization indicated influence AD = 7.9 −4 bipolar disorder 1.35 −2 BAG. These results emphasize polygenic provide insights into relationship selected neurological neuropsychiatric disorders

Language: Английский

---

Optimising brain age estimation through transfer learning: A suite of pre‐trained foundation models for improved performance and generalisability in a clinical setting

Human Brain Mapping, Journal Year: 2024, Volume and Issue: 45(4)

Published: March 1, 2024

Abstract Estimated age from brain MRI data has emerged as a promising biomarker of neurological health. However, the absence large, diverse, and clinically representative training datasets, along with complexity managing heterogeneous data, presents significant barriers to development accurate generalisable models appropriate for clinical use. Here, we present deep learning framework trained on routine ( N up 18,890, range 18–96 years). We five separate prediction (all mean absolute error ≤4.0 years, R 2 ≥ .86) across different sequences (T ‐weighted, T ‐FLAIR, 1 diffusion‐weighted, gradient‐recalled echo *‐weighted). Our offer dual functionality. First, they have potential be directly employed data. Second, can used foundation further refinement accommodate other (and therefore scenarios which employ such sequences). This adaptation process, enabled by transfer learning, proved effective in our study scan orientations, including those differed considerably original datasets. Crucially, findings suggest that this approach remains viable even limited availability (as low = 25 fine‐tuning), thus broadening application estimation more diverse contexts patient populations. By making these publicly available, aim provide scientific community versatile toolkit, promoting research related areas.

Language: Английский

---

A perspective on brain-age estimation and its clinical promise

Nature Computational Science, Journal Year: 2024, Volume and Issue: 4(10), P. 744 - 751

Published: July 24, 2024

Language: Английский

---

Assessing brain involvement in Fabry disease with deep learning and the brain‐age paradigm

Human Brain Mapping, Journal Year: 2024, Volume and Issue: 45(5)

Published: March 23, 2024

Abstract While neurological manifestations are core features of Fabry disease (FD), quantitative neuroimaging biomarkers allowing to measure brain involvement lacking. We used deep learning and the brain‐age paradigm assess whether FD patients' brains appear older than normal validate brain‐predicted age difference (brain‐PAD) as a possible severity biomarker. MRI scans patients healthy controls (HCs) from single Institution were, retrospectively, studied. The stabilization index (FASTEX) was recorded severity. Using minimally preprocessed 3D T1‐weighted subjects eight publicly available sources ( N = 2160; mean 33 years [range 4–86]), we trained model predicting chronological based on DenseNet architecture it generate predictions in internal cohort. Within linear modeling framework, brain‐PAD tested for age/sex‐adjusted associations with diagnostic group (FD vs. HC), FASTEX score, both global voxel‐level measures. studied 52 (40.6 ± 12.6 years; 28F) 58 HC (38.4 13.4 28F). achieved accurate out‐of‐sample performance (mean absolute error 4.01 years, R 2 .90). had significantly higher (estimated marginal means: 3.1 −0.1, p .01). Brain‐PAD associated score B 0.10, .02), parenchymal fraction −153.50, .001), white matter hyperintensities load 0.85, .01), tissue volume reduction throughout brain. demonstrated that normal. correlates FD‐related multi‐organ damage is influenced by hyperintensities, offering comprehensive biomarker (neurological)

Language: Английский

---

Multiscale functional connectivity patterns of the aging brain learned from harmonized rsfMRI data of the multi-cohort iSTAGING study

NeuroImage, Journal Year: 2023, Volume and Issue: 269, P. 119911 - 119911

Published: Jan. 30, 2023

To learn multiscale functional connectivity patterns of the aging brain, we built a brain age prediction model measures at seven scales on large fMRI dataset, consisting resting-state scans 4186 individuals with wide range (22 to 97 years, an average 63) from five cohorts. We computed individual subjects using personalized network computational method, harmonized multiple datasets in order build model, and finally evaluated how gap correlated cognitive subjects. Our study has revealed that were more informative than those any single scale for prediction, data harmonization significantly improved performance, measures' tangent space worked better their original space. Moreover, scores derived clinical measures. Overall, these results demonstrated learned large-scale multi-site rsfMRI dataset characterizing was associated

Language: Английский

---

Constructing personalized characterizations of structural brain aberrations in patients with dementia using explainable artificial intelligence

npj Digital Medicine, Journal Year: 2024, Volume and Issue: 7(1)

Published: May 2, 2024

Deep learning approaches for clinical predictions based on magnetic resonance imaging data have shown great promise as a translational technology diagnosis and prognosis in neurological disorders, but its impact has been limited. This is partially attributed to the opaqueness of deep models, causing insufficient understanding what underlies their decisions. To overcome this, we trained convolutional neural networks structural brain scans differentiate dementia patients from healthy controls, applied layerwise relevance propagation procure individual-level explanations model predictions. Through extensive validations demonstrate that deviations recognized by corroborate existing knowledge aberrations dementia. By employing explainable classifier longitudinal dataset with mild cognitive impairment, show spatially rich complement prediction when forecasting transition help characterize biological manifestation disease individual brain. Overall, our work exemplifies potential artificial intelligence precision medicine.

Language: Английский

---