Dual Graph Attention based Disentanglement Multiple Instance Learning for Brain Age Estimation

TL;DR

The proposed DGA-DMIL framework demonstrates exceptional accuracy in estimating brain age, achieving a remarkable mean absolute error of 2.12 years in the UK Biobank, and is established as state-of-the-art compared to other competing brain age estimation models.

Abstract

Deep learning techniques have demonstrated great potential for accurately estimating brain age by analyzing Magnetic Resonance Imaging (MRI) data from healthy individuals. However, current methods for brain age estimation often directly utilize whole input images, overlooking two important considerations: 1) the heterogeneous nature of brain aging, where different brain regions may degenerate at different rates, and 2) the existence of age-independent redundancies in brain structure. To overcome these limitations, we propose a Dual Graph Attention based Disentanglement Multi-instance Learning (DGA-DMIL) framework for improving brain age estimation. Specifically, the 3D MRI data, treated as a bag of instances, is fed into a 2D convolutional neural network backbone, to capture the unique aging patterns in MRI. A dual graph attention aggregator is then proposed to learn the backbone features by exploiting the intra- and inter-instance relationships. Furthermore, a disentanglement branch is introduced to separate age-related features from age-independent structural representations to ameliorate the interference of redundant information on age prediction. To verify the effectiveness of the proposed framework, we evaluate it on two datasets, UK Biobank and ADNI, containing a total of 35,388 healthy individuals. Our proposed model demonstrates exceptional accuracy in estimating brain age, achieving a remarkable mean absolute error of 2.12 years in the UK Biobank. The results establish our approach as state-of-the-art compared to other competing brain age estimation models. In addition, the instance contribution scores identify the varied importance of brain areas for aging prediction, which provides deeper insights into the understanding of brain aging.

Figures (9)

• Figure 1: The framework of the proposed DGA-DMIL. A bag of instances is fed into the proposed DGA-DMIL network for predicting age accurately. The essence of DGA-DMIL lies in two key components: (1) Dual Graph Attention Aggregator, responsible for aggregating both intra-instance and inter-instance features (The green box indicates the spatial aggregator $A_{\theta, S}$ and the blue box indicates the instance aggregator $A_{\theta, I}$), and (2) Disentanglement Branch, which separates the backbone features into age-related and age-irrelevant features. (Blue arrows symbolize age-related branches, whereas green arrows signify decoupled branches).
• Figure 2: The backbone of the convolutional neural network is responsible for taking a brain image as input and converting it into a deep feature representation. The network comprises 10 blocks, each consisting of a convolutional layer, a batch normalization layer, and a ReLU layer. The spatial resolution is reduced through the max-pooling layer.
• Figure 3: On the left, our model employs an attention mechanism, characterized by a fully connected layer and a weight vector $a^k$, incorporating a LeakyReLU activation and SoftMax function. On the right, an illustration depicts multi-head attention, featuring K = 3 heads, executed by node 1 within its neighborhood. Various arrow styles and colors signify distinct attention computations. The features from each head are combined, concatenated, and averaged to derive $\tilde{h}_1$.
• Figure 4: The illustration of the Dual Graph Attention Aggregator Block: The aggregators $A_{\theta, S}$ specialize in aggregating spatial information from the deep feature map within an instance. Subsequently, $A_{\theta, I}$ is designed to aggregate instance information from the feature set within a bag. The Dual Graph Attention Aggregator aims to obtain a representative for brain age estimation. In addition, instance scores are output to evaluate the impact of individual instances on brain age estimation.
• Figure 5: Performance of brain age estimation with different parameters: (a) numbers of slices in an instance; (b) numbers of instances in a bag.