Enhancing Brain Age Estimation with a Multimodal 3D CNN Approach Combining Structural MRI and AI-Synthesized Cerebral Blood Volume Data

2024-12-01

arXiv.org

AI Summary

To address insufficient brain age estimation accuracy amid global population aging, this study proposes the first multimodal 3D CNN framework integrating T1-weighted MRI and AI-synthesized contrast-free cerebral blood volume (AICBV). We innovatively introduce AICBV as a functional aging surrogate biomarker and design a structural–functional dual-stream VGG architecture with linear fusion, augmented by Grad-CAM for interpretable localization. Evaluated on a test set of n = 288 subjects, the model achieves a mean absolute error of 3.95 years and an R² of 0.943—significantly outperforming unimodal baselines. Visualization confirms precise identification of key aging-sensitive regions, including the hippocampus and prefrontal cortex. This work establishes a novel paradigm for early neurodegenerative disease detection through synergistic structural–functional neuroimaging biomarkers.

Technology Category

• Computer Vision: Multi-modal Vision
• Humans and AI: Brain-Sensing and Analysis
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

The increasing global aging population necessitates improved methods to assess brain aging and its related neurodegenerative changes. Brain Age Gap Estimation (BrainAGE) offers a neuroimaging biomarker for understanding these changes by predicting brain age from MRI scans. Current approaches primarily use T1-weighted magnetic resonance imaging (T1w MRI) data, capturing only structural brain information. To address this limitation, AI-generated Cerebral Blood Volume (AICBV) data, synthesized from non-contrast MRI scans, offers functional insights by revealing subtle blood-tissue contrasts otherwise undetectable in standard imaging. We integrated AICBV with T1w MRI to predict brain age, combining both structural and functional metrics. We developed a deep learning model using a VGG-based architecture for both modalities and combined their predictions using linear regression. Our model achieved a mean absolute error (MAE) of 3.95 years and an $R^2$ of 0.943 on the test set ($n = 288$), outperforming existing models trained on similar data. We have further created gradient-based class activation maps (Grad-CAM) to visualize the regions of the brain that most influenced the model's predictions, providing interpretable insights into the structural and functional contributors to brain aging.

Problem

• Brain Age Assessment
• Neurodegeneration
• Global Aging Population

Innovation

• Multimodal 3D Convolutional Neural Networks
• AI-generated Cerebral Blood Volume (AICBV)
• Brain Age Prediction