A Brain Age Prediction Method Using Multi-Scale Attention Mechanism and Fully Convolutional Network
Poster No:
159
Submission Type:
Abstract Submission
Authors:
Zhewei Zhang1, Jinping Xu1, Fan Xinxin1
Institutions:
1Shenzhen Institute of Advanced Technology, Shenzhen, Guangdong
First Author:
Co-Author(s):
Introduction:
Currently, deep learning has made significant progress in the field of biomedical image processing[6], and there are some structural MRI-based methods in the field of brain age prediction, but most of the methods used use 2D slices or use 3D convolution for feature extraction followed by age regression[1], but there are some problems as follows: 1) the methods using 2D slices do not focus on the information in 3D voxel space, and 2) the methods using 3D methods do not focus on both global and local information. Therefore, we provide a 3D network that takes into account both local and global information, we use SFCN [3] as the backbone network for feature extraction, and then use 3D cross attention mechanism[2] to fuse global and local features (Figure 1).We collected a total of 2559 cases of data from 4 datasets (ADNI[7], OASIS[8], IXI, CORR[9]), and the results show that our network is able to predict age with high accuracy on different datasets, achieving better results compared to some existing networks. The improved accuracy of our results in predicting brain age may help clinicians in diagnosing diseases and making treatment recommendations.
Methods:
In this paper, to solve the problem of considering both global contextual information and local structural information in 3DMRI, we first used SFCN for feature extraction, then we provide a 3D cross-attention mechanism, which extracts features from the whole 3DMRI image through a global path to obtain global contextual information, and extracts local features through multiple segmentation of local patches to obtain local detailed features, and uses attention to fuse global and local features using the attention mechanism. The study shows that by using the attention module, irrelevant information in contextual features can be ignored by global features and important information can be better extracted from the feature space[5]. At the same time, this cross-attention mechanism, unlike the normal transformer[4], does not use concat to fuse local and global features and does not require spatial alignment.
Results:
In our recent experiments, we compared the age prediction results of six methods on four datasets as well as the combined dataset, and the results show that we achieved the best results on the ADNI, OASIS, and combined datasets. Since the ADNI and OASIS datasets are more focused on older adults, the results show that our network can better predict the age of the brain in older adults. The results also show that our network can better predict the age of the brain in the elderly, and can be used in clinical applications to assist physicians in making targeted diagnoses for the elderly.
Conclusions:
In this work, we proposed a 3D cross-attentional mechanisms with SFCN for brain age prediction from whole-brain sMRI image information, and validated on 4 datasets. Moreover, our methods were superior to several existing methods.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
Keywords:
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
[2] He, Sheng, P. Ellen Grant, and Yangming Ou. "Global-local transformer for brain age estimation." IEEE transactions on medical imaging 41.1 (2021): 213-224.
[3] Peng, Han, et al. "Accurate brain age prediction with lightweight deep neural networks." Medical image analysis 68 (2021): 101871.
[4] Dosovitskiy, Alexey, et al. "An image is worth 16x16 words: Transformers for image recognition at scale." arXiv preprint arXiv:2010.11929 (2020).
[5] He S, Feng Y, Grant P E, et al. Deep relation learning for regression and its application to brain age estimation[J]. IEEE transactions on medical imaging, 2022, 41(9): 2304-2317
[6] Baecker L, Garcia-Dias R, Vieira S, et al. Machine learning for brain age prediction: Introduction to methods and clinical applications[J]. EBioMedicine, 2021, 72: 103600.
[7] Jack Jr C R, Bernstein M A, Fox N C, et al. The Alzheimer's disease neuroimaging initiative (ADNI): MRI methods[J]. Journal of Magnetic Resonance Imaging: An Official Journal of the International Society for Magnetic Resonance in Medicine, 2008, 27(4): 685-691
[8] LaMontagne, Pamela J., et al. "OASIS-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and Alzheimer disease." MedRxiv (2019): 2019-12.
[9] Zuo X N, Anderson J S, Bellec P, et al. An open science resource for establishing reliability and reproducibility in functional connectomics[J]. Scientific data, 2014, 1(1): 1-13.