Abstract:Brain age is a powerful marker of general brain health. Furthermore, brain age models are trained on large datasets, thus giving them a potential advantage in predicting specific outcomes - much like the success of finetuning large language models for specific applications. However, it is also well-accepted in machine learning that models trained to directly predict specific outcomes (i.e., direct models) often perform better than those trained on surrogate outcomes. Therefore, despite their much larger training data, it is unclear whether brain age models outperform direct models in predicting specific brain health outcomes. Here, we compare large-scale brain age models and direct models for predicting specific health outcomes in the context of Alzheimer's Disease (AD) dementia. Using anatomical T1 scans from three continents (N = 1,848), we find that direct models outperform brain age models without finetuning. Finetuned brain age models yielded similar performance as direct models, but importantly, did not outperform direct models although the brain age models were pretrained on 1000 times more data than the direct models: N = 53,542 vs N = 50. Overall, our results do not discount brain age as a useful marker of general brain health. However, in this era of large-scale brain age models, our results suggest that small-scale, targeted approaches for extracting specific brain health markers still hold significant value.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to evaluate the effectiveness of Brain Age as a specific brain health marker. Specifically, the researchers compared the performance of large - scale Brain Age models and direct models in predicting specific health outcomes, especially in the two tasks of Alzheimer's disease (AD) dementia classification and mild cognitive impairment (MCI) progression prediction. The main question of the study was to determine whether using large - scale pre - trained Brain Age models could outperform models trained directly for specific health outcomes, even though these Brain Age models were pre - trained on thousands of times more data than the direct models. ### Research Background - **Brain Age**: Brain Age is a powerful brain health marker, calculated by predicting an individual's actual age from brain imaging data through a machine - learning model. The Brain Age Gap (BAG), that is, the difference between the predicted age and the actual age, can be used as a sign of accelerated aging and development. - **Direct Model vs. Brain Age Model**: Although Brain Age models are trained on large - scale data, a recognized principle in machine learning is that models that directly predict the target variable usually perform better than models that predict proxy variables. Therefore, the researchers hoped to experimentally verify the applicability of this principle in brain health prediction. ### Research Objectives - **Compare Model Performance**: The researchers used datasets from three different continents (N = 1,848) to compare the performance of large - scale Brain Age models and direct models in two tasks: 1. **AD Classification Task**: Predict whether participants have Alzheimer's disease (AD) dementia. 2. **MCI Progression Prediction Task**: Predict whether patients with mild cognitive impairment (MCI) will progress to AD dementia within three years. ### Main Findings - **AD Classification Task**: - The direct model and the fine - tuned Brainage64D model performed the best, and there was no significant difference between them. - The BAG and BAG - finetune models performed the worst, and there was no significant difference between them. - Even in the case of a small sample size (N = 50), using pre - trained Brain Age models did not significantly improve the performance of AD classification. - **MCI Progression Prediction Task**: - The direct model (Direct - AD2prog) and the fine - tuned Brainage64D model (Brainage64D - finetune - AD2prog) both outperformed the direct model (Direct). - There was no significant difference between the two models. - Even in the case of a small sample size, using pre - trained Brain Age models did not significantly improve the performance of MCI progression prediction. ### Conclusions - **Main Conclusion**: Although Brain Age is a useful brain health marker, in the prediction of specific health outcomes, the performance of direct models is not inferior to that of large - scale pre - trained Brain Age models. This indicates that in some cases, small - scale, targeted methods still have important value. - **Future Directions**: The research results do not deny transfer learning as a strategy to improve prediction, nor do they deny Brain Age as a powerful brain health marker. However, considering the computational requirements for training and fine - tuning large models, the research results suggest that in some tasks, small - scale, targeted methods may be more effective.

Evaluation of Brain Age as a Specific Marker of Brain Health

Brain Age Prediction: Deep Models Need a Hand to Generalize

Analysis of Brain Age Gap across Subject Cohorts and Prediction Model Architectures

Brain-age prediction: Systematic evaluation of site effects, and sample age range and size

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

Connectome-based Predictive Modelling of Ageing, Overall Cognitive Functioning and Memory Performance.

Brain Age Prediction with 3D ResNet34 Model in Healthy Control, Mild Cognitive Impairment, and Alzheimer's Disease

Beyond Accuracy: Refining Brain-Age Models for Enhanced Disease Detection

Does pre-training on brain-related tasks results in better deep-learning-based brain age biomarkers?

The Link between Brain age and Alzheimer’s Disease development using different neuroimaging modalities

MRI Signatures of Brain Age in the Alzheimer’s Disease Continuum

Rate of brain aging associates with future executive function in Asian children and older adults

Brain Age Estimation on a Dementia Cohort Using FLAIR MRI Biomarkers

Estimating brain age based on a healthy population with deep learning and structural MRI

Brain age as a biomarker for pathological versus healthy ageing – a REMEMBER study

Brain age prediction: A comparison between machine learning models using region‐ and voxel‐based morphometric data

Machine Learning-Driven Prediction of Brain Age for Alzheimer's Risk: APOE4 Genotype and Gender Effects

Gray Matter Age Prediction as a Biomarker for Risk of Dementia

Brain age identification from diffusion MRI synergistically predicts neurodegenerative disease

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

BrainAgeNeXt: Advancing Brain Age Modeling for Individuals with Multiple Sclerosis