Do Transformers and CNNs Learn Different Concepts of Brain Age?
Nys Tjade Siegel1, Dagmar Kainmueller2,3,4, Fatma Deniz5,6
1Department of Psychiatry and Neurosciences, Charité - Universitätsmedizin Berlin (Corporate Member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health), Berlin, Germany.
Human Brain Mapping
|June 9, 2025
Summary
Transformers and convolutional neural networks show similar performance in predicting brain age from MRI scans. Different deep learning models capture comparable brain aging effects, ensuring reliable biomarker analysis.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Biomarker Development
Background:
- Predicted brain age is a biomarker of brain health using machine learning on MRI.
- Convolutional Neural Networks (CNNs) are current state-of-the-art for brain age prediction.
- Transformers represent a newer deep learning architecture with potential for improved performance.
Purpose of the Study:
- To compare the performance of transformer models against CNNs for brain age prediction.
- To investigate if different deep learning architectures capture distinct aspects of brain aging.
- To analyze the practical implications and clinical utility of brain age predictions from various models.
Main Methods:
- Adapted Simple Vision Transformer (sViT) and shifted window transformer (SwinT) models.
- Compared transformer models with a ResNet50 (CNN) architecture.
- Utilized 46,381 T1-weighted structural MR images from the UK Biobank dataset.
Main Results:
- SwinT and ResNet50 demonstrated comparable brain age prediction accuracy.
- Transformer models (sViT, SwinT) and ResNet50 captured similar patterns of brain aging.
- No substantial differences were observed in the structure of brain age predictions across architectures.
Conclusions:
- Deep learning model architecture does not appear to confound brain age studies.
- Transformers show promise for brain age prediction, with potential for further improvement.
- The choice of model architecture does not significantly alter the interpretation of brain age biomarkers.
Related Concept Videos
01:01Neuroplasticity
309
Neuroplasticity reflects the brain's remarkable capacity to adapt and evolve, responding dynamically to learning, experiences, or injury by reorganizing its neural circuitry. This reorganization involves creating new neural connections and refining old ones through a series of biological processes that contribute to the brain's lifelong development and adaptability.
309
01:24Concepts and Prototypes
117
The human nervous system handles vast amounts of information by translating sensory stimuli into neural impulses, which the brain processes, creating thoughts expressed through language or stored as memories. The brain also synthesizes information from emotions and memories, which significantly influence thoughts and behaviors. This intricate process creates a comprehensive mental picture.
The brain organizes this information using concepts, which are mental categories grouping linguistic data,...
The brain organizes this information using concepts, which are mental categories grouping linguistic data,...
117
01:20Introduction to Cognitive Psychology
448
Cognitive psychology is the field of psychology dedicated to examining how people think. It attempts to explain how and why we think the way we do by studying the interactions among human thinking, emotion, creativity, language, and problem-solving, as well as other cognitive processes. Cognitive psychology studies how information is processed and manipulated in remembering, thinking, and knowing.
This field emerged in the mid-20th century, following a period dominated by behaviorism, which...
This field emerged in the mid-20th century, following a period dominated by behaviorism, which...
448