Abstract
Age is one of the major known risk factors for Alzheimer’s Disease (AD). Detecting AD early is crucial for effective treatment and preventing irreversible brain damage. Brain age, a measure derived from brain imaging reflecting structural changes due to aging, may have the potential to identify AD onset, assess disease risk, and plan targeted interventions. Deep learning-based regression techniques to predict brain age from magnetic resonance imaging (MRI) scans have shown great accuracy recently. However, these methods are subject to an inherent regression to the mean effect, which causes a systematic bias resulting in an overestimation of brain age in young subjects and underestimation in old subjects. This weakens the reliability of predicted brain age as a valid biomarker for downstream clinical applications. Here, we reformulate the brain age prediction task from regression to classification to address the issue of systematic bias. Recognizing the importance of preserving ordinal information from ages to understand aging trajectory and monitor aging longitudinally, we propose a novel ORdinal Distance Encoded Regularization (ORDER) loss that incorporates the order of age labels, enhancing the model’s ability to capture age-related patterns. Extensive experiments and ablation studies demonstrate that this framework reduces systematic bias, outperforms state-of-art methods by statistically significant margins, and can better capture subtle differences between clinical groups in an independent AD dataset. Our implementation is publicly available at https://github.com/jaygshah/Robust-Brain-Age-Prediction.
Introduction
Brain age prediction—estimating biological age from MRI scans—is a powerful tool for spotting neurodegeneration. However, current deep learning models, primarily based on regression, suffer from a "regression to the mean" bias. This systematic bias causes models to overestimate the age of young subjects and underestimate the age of older subjects, potentially masking early signs of atrophy in clinical settings. The proposed framework addresses this by reframing the task as an ordinal classification problem.
Methods: Reframing Age Prediction
The authors propose shifting from regression to Ordinal Classification using a 3D ResNet-18 architecture. Unlike standard classification, which treats age labels as independent categories, this approach preserves the natural order of aging by calculating the final age as a weighted average of class probabilities.
ORdinal Distance Encoded Regularization (ORDER) Loss
The core innovation is the ORDER loss, a regularization term added to the standard Cross-Entropy loss. It forces the model's internal "embedding space" to reflect chronological distances. Specifically, it ensures that the mathematical distance between two brain scan representations is proportional to the actual age difference between the subjects.
The authors found that using Manhattan distance ($L_1$ norm) for feature separation significantly outperformed Euclidean distance, mitigating the "curse of dimensionality" where high-dimensional points appear equidistant in standard Euclidean space.
Datasets and Experimental Setup
The study utilized a massive, heterogeneous dataset to ensure model robustness:
- Lifespan Cohort (Training/Validation): 7,377 T1-weighted MRI scans from five public sources (NACC, OASIS, ICBM, IXI, ABIDE), spanning ages 8 to 95.
- Discovery Cohort (Independent Testing): 1,584 scans from the ADNI database, used to evaluate clinical sensitivity across healthy controls, Mild Cognitive Impairment (MCI), and Alzheimer’s patients.
Key Results
The ORDER framework demonstrated significant improvements over state-of-the-art regression and classification methods:
- Accuracy: Achieved a Mean Absolute Error (MAE) of 2.56 years, compared to 3.93 years for standard regression.
- Bias Reduction: Systematic bias in young subjects dropped from 3.4 years to just 0.1 years, effectively solving the regression-to-the-mean problem.
- Clinical Sensitivity: The model effectively captured the "Brain Age Gap" (Predicted Age - Chronological Age). It successfully distinguished healthy controls from those who would later convert to MCI ($p=0.07$), a distinction standard regression models failed to make ($p=0.34$).
Conclusion
By reframing brain age prediction as an ordinal classification task and introducing distance-based regularization, this work effectively eliminates systematic bias. This makes brain age a more reliable biomarker for early Alzheimer's detection and longitudinal monitoring of neurodegeneration. Future work may extend this ORDER loss to other medical regression tasks where preserving ordinality is critical.
Summary updated